<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Anthony Bahn on Medium]]></title>
        <description><![CDATA[Stories by Anthony Bahn on Medium]]></description>
        <link>https://medium.com/@anthonymbahn?source=rss-a927b97facd9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*WCeePc3ovYNinNKo</url>
            <title>Stories by Anthony Bahn on Medium</title>
            <link>https://medium.com/@anthonymbahn?source=rss-a927b97facd9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 08:02:46 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@anthonymbahn/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How to Hot-Patch UnrealIRCd and Verify the Fix]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-hot-patch-unrealircd-and-verify-the-fix-b7e511ea7819?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/b7e511ea7819</guid>
            <category><![CDATA[patch-management]]></category>
            <category><![CDATA[irc]]></category>
            <category><![CDATA[vulnerability]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[unrealircd]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 14:39:41 GMT</pubDate>
            <atom:updated>2026-11-15T14:39:41.755Z</atom:updated>
            <content:encoded><![CDATA[<p>To hot-patch UnrealIRCd for CVE-2026–114713, apply the network-wide SPAMFILTER workaround, run ./unrealircd hot-patch bundle-20261107 on every server from 6.0.0 to 6.2.7, then confirm each one with module-status. The hot-patch fixes the flaw without a restart, so users stay connected. Servers that cannot take the hot-patch upgrade to 6.2.7.1, and UnrealIRCd 5 servers blacklist the labeled-response module and restart.</p><h3>What is an UnrealIRCd hot-patch?</h3><p>An UnrealIRCd hot-patch is a bundle of fixed modules that the server builds and loads while it is running, then activates with a rehash. It changes the code of the affected modules without disconnecting users, which a full upgrade and restart would do.</p><p>The project published bundle-20261107 on November 7, 2026, alongside UnrealIRCd 6.2.7.1. It covers all nine security issues fixed in that release, including <a href="https://www.anthonybahn.com/news/unrealircd-cve-2026-114713-exploited-cisa-kev/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">CVE-2026–114713, which CISA added to its exploited list</a> on November 13. The commands below come from the maintainer’s oss-security advisory.</p><h3>Before you start</h3><ul><li><strong>List every server.</strong> The flaw abuses links between servers, so a single unpatched leaf leaves the network exposed. Include hubs, leaves and any services or test servers that link in.</li><li><strong>Record versions.</strong> Note which servers run 6.0.0 to 6.2.7, which run 5.x and which run older releases. The hot-patch covers only 6.0.0 to 6.2.7.</li><li><strong>Check how each server was installed.</strong> The commands below run from the UnrealIRCd directory of a source install. Distribution packages, such as unrealircd-6.1.10–1.el9 in EPEL 9, follow the package manager’s update path instead. As of November 9, Red Hat’s tracking bugs for the package fix were still open.</li><li><strong>Have IRC operator access.</strong> You need it for the workaround and for checking remote servers.</li></ul><h3>Step 1: apply the SPAMFILTER workaround</h3><p>The advisory gives a one-line SPAMFILTER command, run once as an IRC operator, that blocks label tags containing invalid characters. The advisory presents it as a network-wide workaround, which covers linked servers while you patch them one at a time. Copy the command exactly from the advisory: it contains a <a href="https://www.anthonybahn.com/glossary/regular-expression/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">regular expression</a>, and a single changed character breaks it.</p><p>The advisory notes one limit: the workaround protects the network only when all linked servers run 6.0.4 or newer. If any server is older, move it up the list for patching or upgrading.</p><h3>Step 2: hot-patch UnrealIRCd on each server</h3><p>From the UnrealIRCd directory on each 6.x server, run:</p><pre>./unrealircd hot-patch bundle-20261107</pre><p>On success the last line reads “Patch installed and server rehashed correctly. All should be good now!” Older 6.x versions print “Done! All should be good now” instead. Either message means the patched module is loaded.</p><p>If the build fails, check the UnrealIRCd FAQ entry on failed hot-patches. The advisory calls out one known cause: a distribution upgrade on the host since UnrealIRCd was compiled can break the build. In that case, upgrade instead (see below).</p><h3>How do you verify the UnrealIRCd hot-patch worked?</h3><p>Run the module-status command on the server and check the version of the labeled-response module. A patched server shows the hot-patch bundle name or version 6.2.7.1; an unpatched one shows only the old module version.</p><p>On 6.0.2 and later:</p><pre>./unrealircd module-status|grep labeled-response</pre><p>Read the output this way:</p><ul><li>labeled-response 5.0-hot-patch-bundle-20261107: the hot-patch is installed.</li><li>A version showing 6.2.7.1: the server runs the fixed release.</li><li>5.0 with no bundle name: the hot-patch did not apply. Repeat step 2 or upgrade instead.</li><li>No output at all: check from IRC instead.</li></ul><p>To check from IRC as an operator, run MODULE -all on the local server and find labeled-response in the list. For another server on the network, use MODULE -all remote.server.name. This is the fastest way to confirm a whole network from one session.</p><h3>Upgrade when the hot-patch is not an option</h3><p>On Unix systems, run the built-in upgrade and then restart the server:</p><pre>./unrealircd upgrade</pre><p>On Windows, or if you prefer a clean install, download 6.2.7.1 from unrealircd.org. A restart disconnects that server’s users, so schedule it per server and let users reconnect to other linked servers in the meantime.</p><h3>How do you protect UnrealIRCd 5 servers?</h3><p>Add a line to the configuration on every UnrealIRCd 5 server that blacklists the labeled-response module, then restart the server. A rehash is not enough, because the module is already loaded until the process restarts.</p><pre>blacklist-module &quot;labeled-response&quot;;</pre><p>UnrealIRCd 5 reached end of life on July 1, 2023, and receives no fixes. The blacklist removes this one attack path. The other eight bugs fixed in 6.2.7.1 may also affect it, and the maintainer did not test 5.x, 4 or 3.2.x for them. Put the <a href="https://www.anthonybahn.com/glossary/end-of-life-eol-software/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">end-of-life</a> servers on a dated plan to move to version 6.</p><h3>Check for signs of misuse</h3><p>The project reported exploitation starting within two days of the release. After patching, review each server’s logs from November 7 onward:</p><ul><li>KILL and GLINE actions that no operator recalls issuing.</li><li>Bans, kills or mode changes attributed to a server rather than a person.</li><li>New or changed operator accounts and server links.</li></ul><p>Anything unexplained means someone exercised server authority on your network. Review operator credentials, link passwords and services accounts, and keep the logs.</p><h3>Quick checklist</h3><ol><li>SPAMFILTER workaround applied once, network-wide.</li><li>Hot-patch run on every 6.0.0 to 6.2.7 server.</li><li>module-status or MODULE -all shows the bundle or 6.2.7.1 everywhere.</li><li>Failed or Windows servers upgraded to 6.2.7.1 and restarted.</li><li>UnrealIRCd 5 servers blacklist labeled-response and were restarted.</li><li>Logs reviewed since November 7.</li></ol><p>For the root cause, see <a href="https://www.anthonybahn.com/learn/what-is-crlf-injection/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">what CRLF injection is and how to prevent it</a>. For why one module bug reached every server, read <a href="https://www.anthonybahn.com/learn/why-linked-server-trust-spreads-compromise/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">why linked server trust turns one bug into a takeover</a>.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/hot-patch-unrealircd-verify-fix/">https://www.anthonybahn.com/learn/hot-patch-unrealircd-verify-fix/</a>.</p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=hot-patch-unrealircd-verify-fix">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b7e511ea7819" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Find World-Writable Sockets on Linux]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-find-world-writable-sockets-on-linux-25290007d77d?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/25290007d77d</guid>
            <category><![CDATA[privilege-escalation]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[system-administration]]></category>
            <category><![CDATA[unix]]></category>
            <category><![CDATA[linux]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 19:21:08 GMT</pubDate>
            <atom:updated>2026-11-14T19:21:08.931Z</atom:updated>
            <content:encoded><![CDATA[<p>Use find and ss to locate world-writable sockets, map each one to its process, and triage root daemons anyone on the host can reach.</p><p>To find world-writable sockets on Linux, search the filesystem for socket files with the “other” write bit set using find, then map each one to the process listening on it with ss. Any socket that is writable by every user and owned by a root process deserves a closer look, because on Linux write permission on a socket file is what allows a process to connect. The whole sweep takes a few minutes per host.</p><h3>Why look for world-writable sockets now</h3><p>The exploited <a href="https://www.anthonybahn.com/news/check-point-gaia-cve-2026-123039-root-exploited-cisa-kev/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=find-world-writable-sockets-linux">Check Point CVE-2026–123039 flaw</a> came down to one socket file: /tmp/dynamic_socket, mode 0777, served by a root daemon that ran any command it received. Check Point&#39;s verification step for the fix is simply checking that the socket is now 0660.</p><p>The same class of mistake turns up in in-house tools, vendor agents and appliance software. A sweep of your own Linux servers is cheap and finds the obvious cases before someone else does. If the socket concept is new to you, start with <a href="https://www.anthonybahn.com/learn/what-is-unix-domain-socket/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=find-world-writable-sockets-linux">what a Unix domain socket is and how its permissions work</a>.</p><h3>What counts as a world-writable socket?</h3><p>A world-writable socket is a socket file whose mode grants write permission to “other”, meaning every account on the host. In a long directory listing its mode starts with the letter s and has a w in the last permission group.</p><p>Examples are srwxrwxrwx (0777) and srw-rw-rw- (0666).</p><h3>How do you find world-writable sockets with find?</h3><p>Use GNU find with a type test for sockets and a permission test for the “other” write bit, run as root across each local filesystem. Then map each match to its listening process.</p><p>In find, -type s matches sockets. The -perm -mode form matches files that have all the listed bits set, so -perm -o+w matches anything writable by other users.</p><h3>Step 1: Sweep local filesystems</h3><p>Run as root so permission errors do not hide results:</p><pre>find / -xdev -type s -perm -o+w</pre><p>-xdev keeps find on the root filesystem and stops it from descending into other mounts. Repeat the command for each separate local mount you care about (for example /var or /run if they are separate filesystems), or drop -xdev and accept a slower run.</p><h3>Step 2: Check the common socket locations</h3><p>Most services put sockets under /run, /var/run or /tmp. Sockets in /tmp deserve the most suspicion: the directory is shared by every user, which is the opposite of what a privileged service needs. Run a targeted listing:</p><pre>find /run /tmp -type s -perm -o+w</pre><h3>Step 3: Map each socket to its process</h3><p>A socket file alone does not tell you what is listening. List listening Unix sockets together with the owning process:</p><pre>ss -xlp</pre><p>-x limits output to Unix sockets, -l to listening ones, and -p shows the process. Match the paths from steps 1 and 2 against this output, and note the user each process runs as.</p><h3>Step 4: Triage the results</h3><p>Sort what you find into three groups:</p><ol><li><strong>Expected and safe.</strong> Some sockets are world-writable by design because the service authenticates every caller itself. Confirm that in the vendor documentation, not by assumption.</li><li><strong>Root process, unclear authentication.</strong> This is the dangerous group. Ask the vendor or the code owner how the daemon decides who may call it. If the answer is “the file mode”, treat it as a finding.</li><li><strong>Unknown process or orphaned socket.</strong> A socket with no listener, or one served by software nobody recognizes, needs investigation. On a server that should be clean, an unexplained root listener in /tmp can be a sign of compromise.</li></ol><h3>How do you fix a world-writable socket?</h3><p>Fix it in the service configuration or code: create the socket with mode 0660 or tighter, owned by a dedicated group, inside a private directory. Changing the mode on the live file is only a stopgap, because most services recreate the socket on restart.</p><p>The daemon should also check each caller’s identity with SO_PEERCRED instead of trusting the file mode alone.</p><p>For vendor appliances you cannot reconfigure, follow the vendor’s fix. In the Check Point case that means installing the fixed Jumbo <a href="https://www.anthonybahn.com/glossary/hotfix/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=find-world-writable-sockets-linux">Hotfix</a> take, or stopping the dynamic_server daemon as sk1000362 describes until you can.</p><h3>Make it part of routine checks</h3><ul><li><strong>Baseline.</strong> Record the expected set of world-writable sockets per server role, so new ones stand out.</li><li><strong>Automate.</strong> Run the find sweep from your configuration management or monitoring agent and alert on any path not in the baseline.</li><li><strong>Review vendor agents.</strong> Backup, monitoring and EDR agents often run as root and expose local sockets. Ask how they authenticate callers during procurement.</li><li><strong>Include appliances.</strong> Firewalls, load balancers and storage appliances run Linux underneath. Where the vendor allows shell access, include them in the sweep.</li></ul><p>A root daemon with an open socket turns any local foothold into full control. On network appliances that is a short path, which is why <a href="https://www.anthonybahn.com/learn/why-local-privilege-escalation-firewalls-urgent/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=find-world-writable-sockets-linux">local privilege escalation on firewalls is urgent</a> to patch.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/find-world-writable-sockets-linux/">https://www.anthonybahn.com/learn/find-world-writable-sockets-linux/</a>.</p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=find-world-writable-sockets-linux">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=find-world-writable-sockets-linux">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=25290007d77d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Browser Zero-Day Response When the Vendor Shares Nothing]]></title>
            <link>https://medium.com/@anthonymbahn/browser-zero-day-response-when-the-vendor-shares-nothing-8f7348a4995b?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/8f7348a4995b</guid>
            <category><![CDATA[google-chrome]]></category>
            <category><![CDATA[zero-day]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[patch-management]]></category>
            <category><![CDATA[incident-response]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 14:44:42 GMT</pubDate>
            <atom:updated>2026-11-13T14:44:42.596Z</atom:updated>
            <content:encoded><![CDATA[<p>Browser <a href="https://www.anthonybahn.com/glossary/zero-day/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">zero-day</a> response starts the moment a vendor says an <a href="https://www.anthonybahn.com/glossary/exploit/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">exploit</a> exists in the wild, usually before any technical detail is public. You will not get indicators, affected sites or attack chains on day one. The job is to confirm exposure, push and verify the fix across every Chromium-based browser, protect the highest-risk users first and hunt on behavior. This playbook uses the November 2026 Chrome XML zero-day, CVE-2026–125350, as the worked example.</p><h3>What does “exploit exists in the wild” mean for browser zero-day response?</h3><p>It means the vendor has evidence that attackers are already using the bug, not just that a proof of concept exists. Google uses that wording in Chrome release notes when a fix ships for an exploited flaw, and it should move the patch to your emergency track.</p><p>For CVE-2026–125350, Google’s November 9 <a href="https://www.anthonybahn.com/glossary/stable-channel/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">Stable channel</a> update carried that sentence and nothing else about the attacks. CISA added the CVE to its KEV catalog the next day. That pattern, vendor confirmation then KEV listing within a day or two, is typical for Chrome zero-days.</p><h3>Step 1: Confirm scope in the first hour</h3><p>Write down the facts you have and nothing more:</p><ul><li><strong>CVE and component.</strong> CVE-2026–125350, use-after-free in XML.</li><li><strong>Fixed version per platform.</strong> Chrome 157.0.8103.50/.51 on Windows and Mac, 157.0.8103.50 on Linux.</li><li><strong>Severity and exploitation status.</strong> Chromium severity High, CISA-ADP CVSS 8.8, exploited per Google.</li><li><strong>Deadline.</strong> KEV due date November 24, 2026.</li></ul><p>Then list every product in your environment that embeds the affected engine. For a Chromium bug that means Chrome, Edge, Opera, Brave, Vivaldi and any Electron or WebView-based applications. CISA’s KEV entry for this bug names Chrome, Edge and Opera explicitly and says other Chromium browsers may be affected. Each one gets its own line in your tracker with its own fixed version, which may not exist yet. Our earlier piece on <a href="https://www.anthonybahn.com/learn/the-chromium-downstream-problem-tracking-patches-in-forks-and-embedded-browsers/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">tracking patches in Chromium forks and embedded browsers</a> covers that inventory in depth.</p><h3>Step 2: Push the fix and force the relaunch</h3><p>Chrome downloads updates in the background, but Google’s help documentation is clear that the update applies when the browser restarts. A user with forty tabs open for a week is still running the old build.</p><p>Push the update through your normal channel and pair it with a forced relaunch window. The policy mechanics are in our guide on <a href="https://www.anthonybahn.com/learn/how-to-enforce-browser-updates-across-a-managed-fleet/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">how to enforce browser updates across a managed fleet</a>. For an exploited zero-day, shorten the relaunch grace period to hours, not days.</p><h3>How do you verify a browser patch actually landed?</h3><p>Verify from inventory data that reports the running browser version per device, then compare it to the fixed build. A deployment console that says “succeeded” only tells you the installer ran, not that the browser restarted.</p><p>Practical checks:</p><ul><li>Pull a report of Chrome versions from your endpoint management or EDR tool and sort by version. Anything below 157.0.8103.50 is still exposed.</li><li>Spot-check a few machines by hand. Google’s help page says to open More, then Help, then About Google Chrome; the version number appears under the heading, and a Relaunch button means an update is pending.</li><li>Repeat the report at 24 and 72 hours. The long tail is usually laptops that were asleep, kiosks and shared machines nobody logs into.</li></ul><h3>Step 3: Protect the people most likely to be targeted</h3><p>Exploited Chrome zero-days are often used against a narrow set of targets before they spread. Google has not said who was targeted with CVE-2026–125350, so make a reasonable assumption: executives, finance staff, administrators with privileged access and anyone who handles sensitive external communications.</p><p>Patch and verify those machines first. If any of them cannot take the update immediately, move their privileged work to a patched machine until they can.</p><h3>Step 4: Hunt on behavior, not indicators</h3><p>With no published indicators, signature-based detection has nothing to match. Look for behavior that a browser exploit followed by a <a href="https://www.anthonybahn.com/glossary/sandbox-escape/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">sandbox escape</a> would produce:</p><ul><li>Browser processes spawning shells, script interpreters or unsigned executables.</li><li>New files written to startup locations or user profile folders shortly after browsing activity.</li><li>Outbound connections from non-browser processes that started right after a browser session.</li></ul><p>Scope the hunt to the window between the earliest likely exploitation date and the time each device was patched. For CVE-2026–125350, the bug was reported to Google on November 6, so start a few days before that.</p><h3>Step 5: Communicate without guessing</h3><p>Tell leadership what you know, what you have done and when you will update them. Avoid filling gaps with speculation about attackers or targets. Vendors restrict details for a reason, explained in <a href="https://www.anthonybahn.com/learn/why-vendors-withhold-zero-day-details/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">why vendors withhold zero-day details</a>, and your report should reflect that uncertainty plainly.</p><p>A useful status update has four lines: the CVE and what it affects, the percentage of devices on the fixed version, the browsers still waiting for a vendor fix, and the next checkpoint time.</p><h3>How long should browser zero-day response take?</h3><p>Aim to have every managed device on the fixed version within 72 hours of the vendor release, with high-risk users done on day one. CISA’s KEV deadline for federal agencies is often two weeks out, but attackers do not wait for it.</p><p>Track two numbers to measure yourself: time from vendor release to 90 percent coverage, and time to 100 percent. The gap between them shows you where your process breaks. For background on the bug class behind this incident, see <a href="https://www.anthonybahn.com/learn/what-is-use-after-free-vulnerability/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">what a use-after-free vulnerability is</a>, and for the incident itself, the <a href="https://www.anthonybahn.com/news/chrome-zero-day-cve-2026-125350-xml-use-after-free/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">Chrome zero-day CVE-2026–125350 news report</a>.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/browser-zero-day-response-no-vendor-details/">https://www.anthonybahn.com/learn/browser-zero-day-response-no-vendor-details/</a>.</p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=browser-zero-day-response-no-vendor-details">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8f7348a4995b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Block Outbound SMB to Stop NTLM Hash Leaks]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-block-outbound-smb-to-stop-ntlm-hash-leaks-2b164f7bdc42?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/2b164f7bdc42</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[windows-security]]></category>
            <category><![CDATA[firewall]]></category>
            <category><![CDATA[ntlm]]></category>
            <category><![CDATA[smb]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Thu, 12 Nov 2026 14:38:15 GMT</pubDate>
            <atom:updated>2026-11-12T14:46:58.633Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*b_5XejdQgcTa_T4i" /></figure><p>To block outbound SMB, deny TCP 445 from your internal networks to the internet at the perimeter <a href="https://www.anthonybahn.com/glossary/firewall/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">firewall</a>, then add a matching host firewall rule on laptops that leave the office. Test first in log-only mode to find the few systems with a real need. This stops most <a href="https://www.anthonybahn.com/glossary/ntlm/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">NTLM</a> hash leaks, including the one used in the November 2026 Outlook <a href="https://www.anthonybahn.com/glossary/zero-day/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">zero-day</a>, CVE-2026–124194, from reaching an attacker’s server.</p><h3>Why block outbound SMB at all?</h3><p>Blocking outbound SMB stops Windows from sending a user’s NTLM authentication to servers on the internet. Most organizations have no business need for workstations to reach external file shares over TCP 445, so the rule costs little and removes a common credential leak path.</p><p>SMB is the protocol Windows uses for file shares. When a file, link or bug tells Windows to open a remote share, it tries to log in with the current user’s credentials. Inside a domain that is convenient. Pointed at an attacker’s server, it hands over an NTLMv2 response that can be cracked or relayed. Our explainer on <a href="https://www.anthonybahn.com/learn/what-is-ntlm-hash-leak/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">what an NTLM hash leak is</a> covers why that response is valuable.</p><p>The <a href="https://www.anthonybahn.com/news/outlook-zero-day-cve-2026-124194-ntlm-hash-leak/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">Outlook zero-day CVE-2026–124194</a> is a current example. A crafted message file made Outlook connect to an attacker-controlled location without its normal warning. With outbound SMB blocked, that connection fails at your edge.</p><h3>How do you find out who needs outbound SMB?</h3><p>You find out by logging before you block. Run the new rule in log-only or alert mode for one to two weeks and review every internal host that tried to open TCP 445 to an external address.</p><p>Expect a short list. Typical legitimate cases are:</p><ul><li>A backup appliance or file transfer job that writes to a partner’s or provider’s share</li><li>A hosted file service that still requires SMB access from specific servers</li><li>Legacy line-of-business software configured with a remote path</li></ul><p>Each of these should come from a known server, not from user workstations, and should go to a known destination. That lets you write a narrow exception (one source, one destination, one port) instead of leaving the whole rule open.</p><p>Workstations that show up in the log with no business reason deserve a second look. A connection from a user laptop to an unfamiliar external address on 445 is exactly the pattern a hash leak produces.</p><h3>Step 1: Block at the perimeter firewall</h3><p>Add an outbound deny rule for TCP 445 from all internal address ranges to any external destination, placed above your general outbound allow rule. Add explicit allow rules above it for the exceptions you documented, scoped to source and destination.</p><p>Turn on logging for the deny rule. Those log entries become your detection feed: every hit is either a misconfigured application or an attempted leak worth investigating.</p><p>If you use a cloud firewall or secure web gateway for remote users, apply the same rule there. A perimeter rule only protects devices whose traffic passes through the perimeter.</p><h3>Step 2: Add a host firewall rule for mobile devices</h3><p>Laptops on home or hotel networks bypass your perimeter entirely, so the perimeter rule alone leaves them exposed. Push a host firewall rule through your endpoint management tool that blocks outbound TCP 445 to addresses outside your internal ranges.</p><p>Scope it carefully. The rule should block 445 to the internet while still allowing it to your internal file servers and domain controllers, whether the device connects over VPN or on site. Define your internal ranges once in the policy and reference them, so the rule stays correct when a subnet is added.</p><p>Deploy to a pilot group first. Confirm mapped drives, home folders and Group Policy still work over VPN before rolling it out to everyone.</p><h3>Step 3: Watch for other paths out</h3><p>SMB is not the only way Windows can send credentials to a remote host. When the WebClient service is running, Windows can try to reach a remote path over HTTP-based WebDAV instead, and that traffic uses ordinary web ports your firewall allows.</p><p>You do not need to block web traffic to deal with this. Review which devices actually need the WebClient service and disable it where they do not, and watch proxy logs for WebDAV requests from workstations to unfamiliar hosts. Treat this as a second layer, not a replacement for the SMB block.</p><h3>Step 4: Verify the block works</h3><p>After rollout, confirm the rule is doing its job:</p><ol><li>From a test workstation on the corporate network, attempt a connection to an external host on TCP 445 and confirm it fails and is logged.</li><li>Repeat from a laptop on an outside network with the host rule applied.</li><li>Confirm internal file shares and VPN-based access still work for the same devices.</li><li>Add the firewall deny log to your monitoring so new hits raise an alert instead of sitting in a log file.</li></ol><h3>What blocking outbound SMB does not fix</h3><p>An egress block stops leaked hashes from leaving the building. It does nothing about an attacker already inside your network who can catch and relay authentication internally. Pair it with SMB and LDAP signing, strong passphrases and a plan to reduce NTLM use.</p><p>It also does not replace patching. The Outlook fix for CVE-2026–124194 ships in Office updates, which run separately from Windows Update. Our guide on <a href="https://www.anthonybahn.com/learn/office-update-patching-separate-from-windows/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">why Office update patching needs its own track</a> covers how to close that gap.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/block-outbound-smb-stop-ntlm-hash-leaks/">https://www.anthonybahn.com/learn/block-outbound-smb-stop-ntlm-hash-leaks/</a>.</p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p><a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">More at anthonybahn.com</a></p><p><a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=block-outbound-smb-stop-ntlm-hash-leaks">More at AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2b164f7bdc42" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Audit Appliance Configuration Changes After an Exploit]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-audit-appliance-configuration-changes-after-an-exploit-a902e64c7284?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/a902e64c7284</guid>
            <category><![CDATA[vulnerability-management]]></category>
            <category><![CDATA[incident-response]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[configuration-management]]></category>
            <category><![CDATA[email-security]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 14:34:38 GMT</pubDate>
            <atom:updated>2026-11-11T14:34:38.057Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nWp_OsdHhcrwNKXs" /></figure><p>To audit appliance configuration changes after an <a href="https://www.anthonybahn.com/glossary/exploit/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">exploit</a>, export the current configuration, compare it against a known-good copy from before the exposure window, and explain every difference. Then check admin accounts, routing, credentials and any fields that feed system commands. This matters for flaws like <a href="https://www.anthonybahn.com/news/barracuda-esg-cve-2026-116606-exploited-cisa-kev/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">Barracuda ESG CVE-2026–116606</a>, where an attacker could restore a crafted configuration without logging in. Patching stops new changes; it does not reverse old ones.</p><h3>Why audit appliance configuration changes after patching?</h3><p>Because a patch closes the <a href="https://www.anthonybahn.com/glossary/vulnerability/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">vulnerability</a> but leaves any attacker changes in place. A hostile admin account, a rerouted mail flow or a planted setting survives a <a href="https://www.anthonybahn.com/glossary/firmware/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">firmware</a> update.</p><p>Configuration attacks are also quiet. Mandiant describes CVE-2026–116606 as an unauthenticated attacker restoring a crafted backup archive that modifies global configuration. That produces no failed logins and may not show in the audit trail as an admin action.</p><h3>What do you need before you start?</h3><p>You need a trusted baseline, the current configuration, and the appliance’s logs for the exposure window. Without a baseline you can still audit, but every finding takes longer to explain.</p><p>Collect:</p><ul><li><strong>A known-good configuration export</strong> dated before the vulnerability was disclosed or before the admin interface was exposed. Check your backup system, change tickets and any configuration management tool that pulls exports.</li><li><strong>The current configuration</strong>, exported through the vendor’s supported method. Do not restore or reboot first.</li><li><strong>Logs</strong>: admin logins, configuration change events, web access logs for the management interface, and <a href="https://www.anthonybahn.com/glossary/firewall/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">firewall</a> logs showing who reached it.</li><li><strong>The exposure window</strong>: from the earliest date the flaw could have been used (for the Barracuda pair, no later than the November 2, 2026 disclosure) to the time the patch and access restrictions were in place.</li></ul><p>Save copies of all of this somewhere the appliance cannot reach before you change anything. If you later find compromise, this is your evidence.</p><h3>How do you compare the configurations?</h3><p>Export both versions in the same format and compare them with a text diff tool, then classify each difference as expected, unexplained or hostile.</p><ol><li>Normalize both exports. Remove timestamps, serial counters and generated values that change on every export, so they do not bury real changes.</li><li>Run the diff and list every changed line.</li><li>Match each change to a change ticket, a firmware update note or a known administrator action.</li><li>Anything left over is unexplained. Treat it as hostile until someone proves otherwise.</li></ol><p>If the vendor’s export is binary or encrypted, compare settings screen by screen instead, starting with the high-risk areas below.</p><h3>Which settings matter most on an email gateway?</h3><p>Focus on settings that give an attacker access, data or persistence. On an email security gateway that means accounts, mail flow and anything that executes.</p><ul><li><strong>Administrator accounts and roles.</strong> New users, changed passwords, changed roles, disabled MFA or new API keys.</li><li><strong>Authentication sources.</strong> LDAP or directory servers, SSO settings, and the bind accounts they use.</li><li><strong>Mail routing.</strong> Relay hosts, smart hosts, destination servers per domain, and allowed relay networks. A changed route can copy or divert mail.</li><li><strong>Forwarding, archiving and journaling.</strong> New destinations for copies of mail.</li><li><strong>Policy exceptions.</strong> Added allowlists, disabled scanning or quarantine bypass rules.</li><li><strong>Management access settings.</strong> Admin interface IP restrictions, enabled remote support, SSH or API access.</li><li><strong>System fields that reach the shell.</strong> Timezone, hostname, NTP servers and similar values. CVE-2026–111560 executed commands placed in the timezone field, so any field holding characters that do not belong there (semicolons, backticks, pipes, $( sequences) is a red flag.</li><li><strong>Scheduled tasks and update settings.</strong> Automatic patching turned off, or update servers changed.</li></ul><p>For firewalls, VPNs and load balancers the list shifts toward rules, NAT, VPN users and certificates, but the method is the same.</p><h3>What should you check in the logs?</h3><p>Look for configuration changes that have no matching admin session, and for management interface traffic from sources you do not recognize.</p><ul><li>Configuration change events with no admin login before them.</li><li>Admin logins from unusual addresses or at unusual times.</li><li>Requests to import, restore or upload endpoints in the web access log.</li><li>Management interface connections from outside your management networks in firewall logs.</li><li>Gaps in logging, which can mean logs were cleared or the service restarted.</li></ul><p>Appliance logs often roll over in days. If yours have, the firewall and any SIEM copy are your remaining record.</p><h3>What do you do if you find hostile changes?</h3><p>Stop treating it as a configuration problem and treat it as a compromise. An attacker who could change configuration on the Barracuda chain could also have reached root.</p><ol><li>Keep the evidence you already collected and contact the vendor’s support team.</li><li>Isolate the appliance’s management interface completely while you investigate.</li><li>Rotate every credential the appliance stores or uses: admin, directory bind, relay and API.</li><li>Decide between restoring a clean configuration and replacing the device. If root access is possible, the vendor may recommend a rebuild or replacement, as Barracuda did for compromised ESG units in 2023.</li></ol><h3>How do you make the next audit faster?</h3><p>Keep the baseline current. Export configuration on a schedule, store it off the device, and record every intentional change in a ticket. Then a post-exploit audit becomes a short diff instead of an investigation.</p><p>Also reduce what an alternate path can reach in the first place. The pattern behind CVE-2026–116606 is explained in <a href="https://www.anthonybahn.com/learn/what-is-alternate-path-authentication-bypass/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">what an alternate path authentication bypass is</a>, and our earlier coverage of the <a href="https://www.anthonybahn.com/news/palo-alto-pan-os-cve-2026-0322-management-interface-rce-exploited/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">PAN-OS management interface exploit</a> shows how often appliance management planes are the target.</p><p><em>Originally published at </em><a href="https://www.anthonybahn.com/learn/audit-appliance-configuration-changes/"><em>https://www.anthonybahn.com/learn/audit-appliance-configuration-changes/</em></a><em> .</em></p><p>Free security one-pager (AMB Digital Group): <a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet</a></p><p>More at anthonybahn.com: <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes</a></p><p>More at AMB Digital Group on Substack: <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes">https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-appliance-configuration-changes</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a902e64c7284" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Secure a Self-Hosted AI Inference Server]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-secure-a-self-hosted-ai-inference-server-b274747494e0?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/b274747494e0</guid>
            <category><![CDATA[api-security]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[vulnerability-management]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[infrastructure-security]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 14:43:58 GMT</pubDate>
            <atom:updated>2026-11-10T14:43:58.316Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*poP2hK29zmq9IhGv" /></figure><p>To secure an AI <a href="https://www.anthonybahn.com/glossary/inference-server/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">inference server</a> you self-host, treat it like any other internet-facing admin system: keep the management API off public networks, require authentication on every request including local ones, strip client-supplied forwarding headers at the proxy, patch on a schedule and audit API keys and backends regularly. Model servers hold GPU workers, cloud credentials and prompt data, which is why the <a href="https://www.anthonybahn.com/news/gpustack-cve-2026-102895-auth-bypass-cisa-kev/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">GPUStack CVE-2026–102895 auth bypass</a> turned into code execution on worker hosts.</p><h3>What makes an AI inference server a high-value target?</h3><p>An AI inference server controls expensive compute and stores credentials for the machines and cloud accounts behind it. An attacker with admin access can run code on GPU workers, read prompts and completions, and use stored cloud keys to launch resources at your expense.</p><p>Many of these platforms also started as tools for one developer on one workstation. Defaults chosen for convenience, such as trusting local requests or leaving debug endpoints open, often survive into shared deployments.</p><h3>Step 1: Put the management plane behind a gate</h3><p>Separate the two things the server does. The inference endpoint that applications call may need to be reachable from your app servers. The admin console and management API almost never need to be reachable from the internet.</p><ul><li>Bind the management interface to a private network or restrict it with <a href="https://www.anthonybahn.com/glossary/firewall/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">firewall</a> rules to an admin subnet.</li><li>Put a VPN or an authenticating gateway (SSO with MFA) in front of the console.</li><li>Keep worker nodes on a network segment that only the cluster manager can reach. GPUStack’s CVE-2026–58658 record describes worker /serveLogs and /debug endpoints that exposed inference logs without authentication in versions through 2.2.1.</li></ul><h3>Step 2: Require authentication for local requests</h3><p>Check every product setting that exempts loopback or “local” callers from authentication and turn the exemption off. In GPUStack the setting is force_auth_localhost; the advisory for CVE-2026-102895 recommends setting it to true on versions you cannot upgrade yet.</p><p>This matters most when a <a href="https://www.anthonybahn.com/glossary/reverse-proxy/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">reverse proxy</a> runs on the same host, because the proxy makes all traffic look local. Our explainer on <a href="https://www.anthonybahn.com/learn/what-is-localhost-authentication-bypass/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">what a localhost authentication bypass is</a> covers why.</p><h3>Step 3: Harden the reverse proxy</h3><p>The proxy in front of the server is part of its authentication boundary.</p><ul><li>Strip any client IP header the application trusts (for GPUStack, X-GPUStack-Real-IP) from incoming requests, then set it yourself if the application needs it.</li><li>Log the real client address at the proxy. When the backend sees only loopback, the proxy log is the only record of who called which endpoint.</li><li>Block or restrict paths you do not use. GPUStack’s SAML advisory notes the /auth/saml/callback route was processed on affected versions even when SAML was not configured.</li></ul><h3>Step 4: Patch on a defined schedule</h3><p>Subscribe to the project’s GitHub security advisories and release feed, assign a named owner and set a patch window that matches your other internet-facing systems. For GPUStack today that means 2.2.2 or later, which covers both CVE-2026–102895 and the SAML bypass CVE-2026–102894.</p><p>Fast-moving open-source AI projects ship often and their advisories can lag the fix by weeks. GPUStack’s 2.2.0 release, which fixed the loopback flaw, predates the CVE record by several months, so teams that upgraded routinely were protected before the <a href="https://www.anthonybahn.com/glossary/vulnerability/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">vulnerability</a> had a number.</p><h3>Step 5: Control API keys and service accounts</h3><p>API keys on model servers tend to be long-lived and broadly scoped.</p><ul><li>Give each application its own key, limited to the models it uses.</li><li>Review admin and full-scope keys monthly and delete any without a named owner.</li><li>When a key may have leaked, revoke it rather than only removing it from code or config; our guide on <a href="https://www.anthonybahn.com/learn/leaked-api-key-revoke-not-delete-from-git/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">revoking a leaked API key instead of deleting it from Git</a> explains why.</li><li>Store cloud provider credentials with the narrowest permissions that still let the platform provision GPUs, and set billing alerts on those accounts.</li></ul><h3>How do you check an inference server for compromise?</h3><p>Start with the objects an attacker would create to keep access: users, admin API keys and custom inference backends that no one on your team created. Then review proxy logs for calls to management endpoints from unfamiliar addresses and check worker hosts for unexpected processes.</p><p>A practical review sequence after an exposure:</p><ol><li>Export or screenshot the current list of users, API keys and backends so you have a record before you change anything.</li><li>Compare each entry against change tickets or team knowledge. The in-the-wild GPUStack case involved attacker-created custom backends.</li><li>Search proxy logs for requests to backend creation and SSO callback paths.</li><li>Rotate worker credentials and stored cloud keys, and review the cloud account for instances you did not launch.</li><li>Rebuild worker hosts that show unexplained processes or outbound connections rather than cleaning them in place.</li></ol><h3>How often should you review inference server security?</h3><p>Review access, keys and exposed endpoints at least monthly, and immediately after any advisory for the platform you run. Fold the server into your normal vulnerability scans and asset inventory so it gets the same attention as your VPN and firewall.</p><p>Most teams find the gap is ownership rather than tooling: the platform was set up by a data science group and nobody in IT tracks its patches. The planning side of that problem is covered in <a href="https://www.anthonybahn.com/learn/why-ai-infrastructure-vulnerability-management/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">why AI infrastructure needs vulnerability management</a>.</p><p><em>Originally published at </em><a href="https://www.anthonybahn.com/learn/secure-self-hosted-ai-inference-server/"><em>https://www.anthonybahn.com/learn/secure-self-hosted-ai-inference-server/</em></a></p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=secure-self-hosted-ai-inference-server">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b274747494e0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Set Refrigeration Setpoint Limits and Alerts]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-set-refrigeration-setpoint-limits-and-alerts-5a00465ccb6f?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/5a00465ccb6f</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[operational-technology]]></category>
            <category><![CDATA[industrial-control-system]]></category>
            <category><![CDATA[infrastructure-security]]></category>
            <category><![CDATA[cold-chain]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 14:30:38 GMT</pubDate>
            <atom:updated>2026-11-09T14:30:38.246Z</atom:updated>
            <content:encoded><![CDATA[<p>Refrigeration <a href="https://www.anthonybahn.com/glossary/setpoint/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">setpoint</a> limits are the minimum and maximum values a controller will accept for a unit’s target temperature, defrost settings and alarms. Set them correctly and a compromised or mistaken change cannot push a cold room far outside its safe range without an alert. The joint US fact sheet issued after the 2026 <a href="https://www.anthonybahn.com/news/refrigeration-controller-attacks-cisa-fbi-cold-chain-warning/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">refrigeration controller attacks</a> tells operators to define reasonable setpoint ranges and alert on deviations. This guide covers how to do that without vendor-specific steps.</p><h3>What is a setpoint limit?</h3><p>A setpoint limit is a bound on the value an operator, or an attacker, can enter as a unit’s target. If a freezer’s setpoint is limited to a range of a few degrees below zero, nobody can set it to 50°C from the controller interface.</p><p>Many commercial refrigeration controllers support limits on setpoints, defrost duration and defrost frequency, and separate high and low temperature alarms. Exactly which settings exist, and where they live, depends on the manufacturer and <a href="https://www.anthonybahn.com/glossary/firmware/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">firmware</a>, so confirm the options in your vendor’s documentation or with your refrigeration contractor.</p><h3>Why do setpoint limits matter for security?</h3><p>Setpoint limits matter because the European attacks worked by changing settings, not by breaking hardware. Attackers forced units into continuous defrost and heating, and temperatures reached 50°C. A unit that refuses out-of-range values, or raises an alarm the moment one appears, turns that change into an early warning.</p><p>Limits do not replace removing internet exposure or changing default passwords. An attacker with full administrative access may be able to change the limits too. Their value is in catching mistakes, slowing an intruder down and making sure someone hears about a dangerous change quickly.</p><h3>How to set refrigeration setpoint limits step by step</h3><h3>1. Build an asset list with the product in each unit</h3><p>List every controlled unit: walk-in coolers, freezers, display cases, blast chillers, pharmacy fridges, refrigerated trailers. For each, record what it stores and who is responsible for that product. A setpoint range only makes sense against a product’s tolerance.</p><p>If you do not yet know which controllers you have, start with an inventory. Our earlier guide on <a href="https://www.anthonybahn.com/learn/how-to-inventory-the-ot-controllers-your-facilities-contractors-put-online/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">how to inventory the OT controllers your facilities contractors put online</a> covers that.</p><h3>2. Define the safe operating range per unit</h3><p>Take the range from the product, not from the equipment. Food safety rules, pharmacy storage requirements and blood bank standards set the limits that matter; the controller’s factory range is usually far wider. Write down three values per unit:</p><ul><li><strong>Target setpoint</strong>: the normal operating temperature.</li><li><strong>Allowed setpoint range</strong>: the narrow band staff may ever need to choose.</li><li><strong>Alarm thresholds</strong>: high and low temperatures at which someone must be told, with a delay that avoids false alarms during normal defrost.</li></ul><p>Get the product owner (store manager, pharmacist, lab lead) to sign off. That record is what you compare against later.</p><h3>3. Set limits on defrost behavior</h3><p>Defrost is where the 2026 attacks did their damage. Record the normal defrost schedule, the maximum defrost duration and the termination temperature for each unit, and ask your contractor to cap them in the controller where the firmware allows. A <a href="https://www.anthonybahn.com/glossary/defrost-cycle/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">defrost cycle</a> that never ends should be impossible, or at least alarmed.</p><h3>4. Configure the limits on the controller</h3><p>Have the refrigeration contractor apply the agreed limits, or do it yourself if you have the access and documentation. Do this in a maintenance window, one unit at a time, and confirm the unit returns to its normal setpoint afterward. Change the controller’s default passwords at the same visit; the fact sheet lists that as a core mitigation, and the Dutch victims were running defaults.</p><h3>5. Add deviation alerts outside the controller</h3><p>Alerts generated by the controller can be disabled by whoever controls the controller. Add at least one alert path that does not depend on it:</p><ul><li>A monitoring platform that polls the units and alarms on readings outside the range.</li><li>Independent wireless or battery data loggers in the highest-value units.</li><li>An alert when a setpoint, alarm threshold or defrost parameter changes, if your monitoring system can see configuration.</li><li>A “no data” alert when a unit stops reporting, which catches the <a href="https://www.anthonybahn.com/learn/what-is-denial-of-view-attack-ot/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">denial-of-view condition</a> the agencies describe.</li></ul><p>Send alerts to a phone that someone answers at 3 a.m. Both European waves happened overnight.</p><h3>6. Back up the configuration</h3><p>Export or record each controller’s configuration once the limits are in place and store a copy offline, as the fact sheet recommends. If a controller is tampered with, a known-good configuration lets the contractor restore it quickly instead of rebuilding from memory.</p><h3>7. Test and review</h3><p>Test each alert path at least once: raise a logger’s threshold briefly, or have the contractor simulate a high-temperature condition. Review the ranges when product mix changes and after any controller replacement or firmware update, since updates can reset settings.</p><h3>How do you know the limits are working?</h3><p>You know the limits work when an out-of-range change either fails at the controller or produces an alert that reaches a person within minutes. Anything else, including an alert that only appears on a dashboard nobody watches, means the control exists on paper only.</p><p>Keep a short log of each test: date, unit, what was tested and how long the alert took to arrive. That record is useful for food safety and pharmacy audits as well as for security reviews.</p><h3>Common mistakes</h3><ul><li><strong>Using the factory range.</strong> Controllers often accept far wider values than any product could tolerate.</li><li><strong>Alarming only inside the controller.</strong> If an attacker owns the controller, its alarms are theirs too.</li><li><strong>No owner for alerts.</strong> A deviation alert that goes to a contractor’s general inbox may not be read until morning.</li><li><strong>Forgetting defrost.</strong> Temperature setpoints get limited, while defrost duration stays open.</li></ul><p>Limits and alerts buy time. What staff do with that time is the subject of <a href="https://www.anthonybahn.com/learn/why-cold-storage-sites-need-manual-fallback-plan/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">why cold storage sites need a manual fallback plan</a>.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/refrigeration-setpoint-limits-deviation-alerts/">https://www.anthonybahn.com/learn/refrigeration-setpoint-limits-deviation-alerts/</a></p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=refrigeration-setpoint-limits-deviation-alerts">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5a00465ccb6f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Enforce a Minimum iOS Version With MDM]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-enforce-a-minimum-ios-version-with-mdm-ccf934bc1095?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/ccf934bc1095</guid>
            <category><![CDATA[mdm]]></category>
            <category><![CDATA[microsoft-intune]]></category>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Sun, 08 Nov 2026 14:32:53 GMT</pubDate>
            <atom:updated>2026-11-08T14:32:53.967Z</atom:updated>
            <content:encoded><![CDATA[<p>To enforce a minimum iOS version, combine two MDM controls: a compliance policy that marks devices below your chosen version as noncompliant and cuts off corporate access, and an enforced software update with a deadline that installs the release for the user. The first protects your data from stale phones. The second shrinks the number of stale phones. Both matter after campaigns like Poisoned Source, which targeted iPhones that had skipped years of updates.</p><h3>Why enforce a minimum iOS version?</h3><p>Enforcing a minimum iOS version keeps devices with known, already patched vulnerabilities away from company email, files and SSO. Attackers keep using n-day exploits because many phones never install the fix.</p><p>In November 2026, CISA added Apple CVE-2024–27834 to its KEV catalog, more than two years after Apple patched it, because an <a href="https://www.anthonybahn.com/glossary/apt29/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">APT29</a>-linked campaign used it against iPhones on iOS 16. Sekoia, which documented the campaign, stated that devices on an up to date operating system were safe. See <a href="https://www.anthonybahn.com/news/apple-cve-2024-27834-cisa-kev-apt29-watering-hole/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">our news coverage of CVE-2024–27834</a> for the details.</p><h3>How do you pick the minimum version?</h3><p>Pick the oldest build that contains every fix you consider mandatory, and raise it after each Apple security release. Start from the newest supported major release, then set the floor at its latest patch, not at the major version number.</p><p>Practical rules for setting the floor:</p><ul><li>Read Apple’s security release notes when a new update ships and check for “may have been exploited” language. Those releases justify a short deadline.</li><li>Check the <a href="https://www.anthonybahn.com/glossary/cisa-kev/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">CISA KEV</a> catalog for Apple entries. A KEV addition, even for an old bug, means attackers are using it now.</li><li>Use the OS build number when Apple ships a fix without changing the version string. Microsoft’s Intune documentation notes that security updates typically change the build number, not the OS version.</li><li>Write the current floor and the date it changes into your mobile device policy, so help desk staff can answer “why am I blocked?” without escalating.</li></ul><h3>Set the compliance rule</h3><p>The compliance rule is what blocks access. In <a href="https://www.anthonybahn.com/glossary/microsoft-intune/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">Microsoft Intune</a>, the iOS/iPadOS compliance policy has a Device Properties section with these settings:</p><ul><li><strong>Minimum OS version:</strong> devices below it are marked noncompliant. Users see a link explaining how to upgrade and regain access once they do.</li><li><strong>Minimum OS build version:</strong> use this when Apple ships a security fix that only changes the build. Intune’s documentation says to enter the supplemental build version for Rapid Security Response updates.</li><li><strong>Maximum OS version</strong> and <strong>Maximum OS build version:</strong> optional, used to hold back a release you have not tested.</li></ul><p>Microsoft lists these settings as supported for iOS 17.0 and later. Check how your tenant reports devices older than that, because those are exactly the phones you most want to catch. If they show as compliant by default, block them some other way, such as a device filter or a report you review weekly.</p><p>Compliance only protects data when something acts on it. Pair the policy with a <a href="https://www.anthonybahn.com/glossary/conditional-access/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">Conditional Access</a> rule in <a href="https://www.anthonybahn.com/glossary/entra-id/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">Microsoft Entra ID</a> (or your identity provider’s equivalent) that requires a compliant device for email, file storage and SSO apps. Without that link, a noncompliant flag is just a report.</p><p>Other MDM platforms (Jamf, Kandji, Workspace ONE and others) offer similar minimum OS rules. The names differ, but the logic is the same: evaluate version, flag noncompliance, gate access.</p><h3>Enforce the update itself</h3><p>Blocking access gets the user’s attention. Installing the update fixes the device. Apple’s platform deployment guide describes enforced software updates through declarative device management:</p><ul><li>The administrator sets a target version and a date and time. Devices try to install the update at that time in their local time zone.</li><li>The device downloads and prepares the update once the declaration arrives. Users see the deadline in Settings and in notifications that get more frequent as it approaches, and Do Not Disturb is ignored in the last 24 hours.</li><li>On iOS 18 and iPadOS 18 or later, notifications can be limited to one hour before the deadline, with a restart countdown.</li><li>If the deadline passes, the device tells the user the update is past due and tries to install within the next hour, retrying when it is powered on and online.</li><li>Status reports show whether the device is waiting, downloading or installing, with error codes for problems such as low battery, low storage or no network.</li></ul><p>Apple also documents enforcing a minimum OS version during Automated Device Enrollment, which requires iOS 17 or iPadOS 17 or later. New and reset devices below the floor are guided through an update before Setup Assistant finishes, so they never reach users stale.</p><h3>How do you handle devices that cannot update?</h3><p>Devices that cannot install your minimum version should lose corporate access and be scheduled for replacement. No setting makes an unsupported phone safe.</p><p>Steps that work in practice:</p><ol><li>Export the noncompliant list after the first deadline and split it into “will update” and “cannot update”.</li><li>Contact the “cannot update” owners with a replacement date, not an open-ended warning.</li><li>Allow web-only access through a browser on a managed laptop in the meantime, if the business needs it.</li><li>Remove the devices from MDM once they are replaced, so they do not linger as exceptions.</li></ol><p>The budget and policy side of this is covered in <a href="https://www.anthonybahn.com/learn/why-unsupported-iphones-need-retirement-schedule/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">why unsupported iPhones belong on a retirement schedule</a>. For background on why old Apple patches keep mattering, read <a href="https://www.anthonybahn.com/learn/what-is-pointer-authentication-apple-pac/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">what Pointer Authentication is and what a PAC bypass does</a>.</p><h3>A rollout checklist</h3><ul><li>Set the floor to the latest patch of your supported major release.</li><li>Create the compliance policy with minimum OS version and, where needed, minimum build.</li><li>Link it to Conditional Access for email, file storage and SSO.</li><li>Declare an enforced update with a deadline a few days out, shorter when Apple flags active exploitation.</li><li>Review the noncompliant report after the deadline and start replacements for devices that cannot update.</li><li>Repeat for every Apple security release.</li></ul><p>Originally published at <a href="https://www.anthonybahn.com/learn/enforce-minimum-ios-version-mdm/">https://www.anthonybahn.com/learn/enforce-minimum-ios-version-mdm/</a></p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=enforce-minimum-ios-version-mdm">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ccf934bc1095" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Audit SSL VPN Accounts for Non-SAML Logins]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-audit-ssl-vpn-accounts-for-non-saml-logins-a86b3a79004e?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/a86b3a79004e</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[identity-management]]></category>
            <category><![CDATA[vpn]]></category>
            <category><![CDATA[authentication]]></category>
            <category><![CDATA[aml]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Sat, 07 Nov 2026 14:33:29 GMT</pubDate>
            <atom:updated>2026-11-07T14:33:29.415Z</atom:updated>
            <content:encoded><![CDATA[<p>To audit <a href="https://www.anthonybahn.com/glossary/ssl-vpn/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">SSL VPN</a> logins for non-SAML access, compare what each account was supposed to use against what the logs say it actually used. Export the appliance’s user and authentication settings, pull successful VPN logins for the longest period you retain, and flag every session by a SAML user that did not come through the identity provider. The exploited SonicWall flaw CVE-2026–19868 makes this check urgent on any SAML-protected VPN.</p><h3>How do you audit SSL VPN logins for non-SAML access?</h3><p>Build a list of accounts that should only use SAML, then search VPN authentication logs for successful logins by those accounts through any other method. Each match is a finding that needs an explanation, even if it turns out to be harmless.</p><p>The steps below use general terms because field names and menus vary by vendor and <a href="https://www.anthonybahn.com/glossary/firmware/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">firmware</a>. Use your vendor’s log reference to map each term to the field your appliance records. For SonicWall context, see our news article on <a href="https://www.anthonybahn.com/news/sonicwall-cve-2026-19868-sslvpn-saml-mfa-bypass-exploited/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">CVE-2026–19868 and the SAML MFA bypass</a>.</p><h3>Step 1: Patch first and preserve what you have</h3><p>Do not hold an upgrade for the audit if the appliance is on an exploited build. Export logs and a configuration backup first, because logs stored only on the appliance may not survive for long, then proceed with the upgrade.</p><p>Keep the exports somewhere the VPN appliance and its administrators cannot alter them. A copy on a separate log server or SIEM is best. Our guide to <a href="https://www.anthonybahn.com/learn/answer-cyber-insurance-siem-logging-question/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">the cyber insurance SIEM and logging question</a> covers what to keep and how to prove it.</p><h3>Step 2: List every authentication source</h3><p>Write down each authentication source configured on the appliance: SAML, local users, LDAP or Active Directory, <a href="https://www.anthonybahn.com/glossary/radius/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">RADIUS</a>, <a href="https://www.anthonybahn.com/glossary/tacacs/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">TACACS+</a> and anything else in use. Note which ones can be used for VPN access, as opposed to administrator login or another service.</p><p>Vendor terminology can hide useful detail here. SonicWall’s 2025 SSLVPN notice, for example, says only users with a password set on the <a href="https://www.anthonybahn.com/glossary/firewall/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">firewall</a>’s management interface count as local users. Auto-generated or locally duplicated LDAP and RADIUS users have no password stored in <a href="https://www.anthonybahn.com/glossary/sonicos/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">SonicOS</a>. A long user list on the firewall may therefore contain far fewer local passwords than it first appears to.</p><h3>Step 3: Build the expected state</h3><p>Create a simple spreadsheet with one row per account that can reach the VPN:</p><ul><li>Username, in every form the appliance recognizes (for example, UPN and SAM account name).</li><li>The method the account is supposed to use.</li><li>Whether a non-SAML credential exists for the account.</li><li>The owner, and the reason for any exception.</li><li>Last successful login and method.</li></ul><p>Include disabled-in-theory accounts, former staff, contractors, vendor support accounts and anything migrated from older hardware. The username forms matter because CVE-2024–12802, an earlier SonicWall <a href="https://www.anthonybahn.com/glossary/mfa-bypass/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">MFA bypass</a>, depended on how UPN and SAM account names were handled.</p><h3>Step 4: Compare logs with the expected state</h3><p>Export successful VPN authentication events and join them to your spreadsheet by username. Normalize usernames first (case, domain prefix, and UPN or SAM form) so one person does not appear as three accounts.</p><p>Flag these results for review:</p><ol><li>A SAML-only account that logged in through any other method.</li><li>A successful VPN session with no matching sign-in on the identity provider side at about the same time.</li><li>An account with a non-SAML credential that has never used it. Nothing happened yet, but the credential should go.</li><li>Logins from new source addresses or countries, or at hours the user does not normally work.</li><li>A burst of failed logins followed by a success on the same account.</li></ol><p>The second check is the most useful one because it does not depend on the appliance recording the method accurately. If the VPN recorded a session and the identity provider has no record of authenticating that user, you have a question to answer.</p><h3>What counts as a suspicious non-SAML login?</h3><p>Any successful login by a SAML-only account through another method should be treated as suspicious until someone can explain it. A documented break-glass test with a named person and a date is an explanation; “probably an old setting” is not.</p><h3>How far back should the audit go?</h3><p>Go back as far as your logs allow. SonicWall has not published a start date for exploitation of CVE-2026–19868, so there is no known safe cutoff, and a short retention window is itself a finding to report.</p><h3>Step 5: Investigate each exception</h3><p>For each unexplained session, work out what happened in order:</p><ol><li>Ask the account owner whether the session was theirs, using a channel other than the possibly compromised account.</li><li>Record the source address, session start and end times, and the internal addresses the session reached.</li><li>Review firewall, directory and endpoint logs for the internal systems that session touched.</li><li>Look for changes on the appliance: new local accounts, newly enabled authentication methods and new VPN authorizations.</li></ol><p>If you cannot rule out misuse, treat the account as compromised. Reset its credentials and end active sessions. SonicWall’s 2025 guidance also calls for rotating exposed credentials such as LDAP bind credentials if an administrator account may have been compromised. Our walkthrough on <a href="https://www.anthonybahn.com/learn/detect-ivanti-connect-secure-exploitation-logs/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">how to detect Ivanti Connect Secure exploitation</a> shows a similar evidence-first approach for another VPN product.</p><h3>Step 6: Remove what you do not need</h3><p>Disable or remove every non-SAML credential that has no documented owner and purpose, and take VPN access away from any account that does not need it. Keep a tested administrative path before you make changes, so a mistake or an identity provider outage does not lock everyone out.</p><p>Then re-test. Use a non-production account to try each remaining route, and confirm that the login either requires your identity provider or fails. If you are unclear on why a single password path counts as a weakness, read <a href="https://www.anthonybahn.com/learn/what-is-cwe-308-single-factor-authentication/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">what CWE-308 single-factor authentication means</a>.</p><h3>Step 7: Turn the audit into an alert</h3><p>A one-time audit goes stale the first time someone adds a user. Forward VPN authentication logs off the appliance and alert on a successful login by a SAML-only account through any other method. Repeat the account comparison quarterly, and after every firmware upgrade or hardware migration. The longer-term decision about these paths is covered in <a href="https://www.anthonybahn.com/learn/why-saml-mfa-fails-vpn-login-paths/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">why SAML MFA fails when other login paths remain</a>.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/audit-ssl-vpn-logins-non-saml-accounts/">https://www.anthonybahn.com/learn/audit-ssl-vpn-logins-non-saml-accounts/</a></p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=audit-ssl-vpn-logins-non-saml-accounts">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a86b3a79004e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Disable SSH Password Authentication Safely]]></title>
            <link>https://medium.com/@anthonymbahn/how-to-disable-ssh-password-authentication-safely-0c5a558ba8af?source=rss-a927b97facd9------2</link>
            <guid isPermaLink="false">https://medium.com/p/0c5a558ba8af</guid>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[authentication]]></category>
            <category><![CDATA[openssh]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[ssh]]></category>
            <dc:creator><![CDATA[Anthony Bahn]]></dc:creator>
            <pubDate>Fri, 06 Nov 2026 15:57:41 GMT</pubDate>
            <atom:updated>2026-11-06T16:00:53.111Z</atom:updated>
            <content:encoded><![CDATA[<p>To disable SSH password authentication safely, confirm key login works first, set PasswordAuthentication no where sshd reads it first, check every Match block, validate the file, restart the daemon and verify the effective setting from a second session. Skipping any of these is how admins lock themselves out or leave passwords on. The <a href="https://www.anthonybahn.com/news/openssh-password-bypass-cve-2026-123158-alpine-musl-exploited/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=disable-ssh-password-authentication-safely">OpenSSH password bypass on Alpine</a> made this an urgent stopgap for musl hosts.</p><h3>What does PasswordAuthentication no change?</h3><p>It tells sshd to stop accepting the SSH password authentication method. The default for this setting is yes, so a config file that never mentions it still allows password logins.</p><p>The <a href="https://man.openbsd.org/sshd_config">sshd_config manual</a> lists KbdInteractiveAuthentication as a separate keyword with its own default of yes. Setting PasswordAuthentication no says nothing about it, so review that keyword as a separate decision while you are in the file.</p><h3>Before you change anything</h3><p>Work through this list on each host:</p><ol><li>Confirm your own key login works from a new connection.</li><li>Find anyone still using passwords. On Alpine, search /var/log/messages for sshd lines reading Accepted password for. Those people, scripts and vendor tools will break when you make the change, so move them to keys first.</li><li>Check that keys can log in to the accounts you depend on. Under Alpine’s PAM-less package, an account with ! in its <a href="https://www.anthonybahn.com/learn/what-is-shadow-password-field-locked-accounts/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=disable-ssh-password-authentication-safely">shadow password field</a> cannot log in with a key either.</li><li>Confirm you have another way in, such as a console or out-of-band session, in case the restart goes wrong.</li><li>Keep your current session open until you have tested a new login. Do not close it to test.</li></ol><h3>Why can a setting in sshd_config be ignored?</h3><p>Because sshd uses the first value it reads for most keywords, a line lower in the file or in a later include can have no effect. A match block can also override the global value for the connections it covers.</p><p>The manual states it plainly: “Unless noted otherwise, for each keyword, the first obtained value will be used.” This has two practical results.</p><p>First, placement matters. A drop-in file only wins if its include is read before any other PasswordAuthentication line, which is why people name these files with a low number such as 00-no-passwords.conf. Older systems may not read the drop-in directory at all: Alpine&#39;s default config reads it from 3.19 onward, and earlier releases ignore it.</p><p>Second, Match blocks can undo your work. PasswordAuthentication is one of the keywords allowed inside a block, and when the criteria are satisfied, the block&#39;s keywords override the global section. A leftover Match block that allows passwords for one user or address range keeps them on for that case, even when the global setting says no.</p><h3>How do you disable SSH password authentication?</h3><p>Add a line setting the option to no in a location sshd reads first, change any existing lines that say yes, including those inside match blocks, then test the configuration and restart the daemon.</p><p>Alpine’s <a href="https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/">advisory</a> gives a sequence for its own systems, run as root:</p><pre>mkdir -p /etc/ssh/sshd_config.d<br>echo &#39;PasswordAuthentication no&#39; &gt; /etc/ssh/sshd_config.d/00-no-passwords.conf<br>sed -i.bak -e &#39;1i PasswordAuthentication no&#39; \<br>  -e &#39;s/^\(\s*PasswordAuthentication[^y#]*\)yes/\1no/i&#39; \<br>  /etc/ssh/sshd_config<br>sshd -t &amp;&amp; rc-service sshd restart<br>sshd -T | grep -i passwordauthentication</pre><p>Here is what each part does:</p><ul><li>The first two commands create the drop-in file for releases that read it.</li><li>The sed command adds the setting at the top of the main file, so it applies even on older releases, and changes yes to no on every matching line, including those inside a Match block. It keeps the original as /etc/ssh/sshd_config.bak.</li><li>sshd -t checks the configuration, and the &amp;&amp; means the restart happens only if the check passes.</li><li>sshd -T prints the effective settings.</li></ul><p>The last command should print passwordauthentication no. On other distributions, the service manager command differs, so use the restart command your init system provides, and check first whether your main file includes the drop-in directory before relying on it.</p><h3>How do you confirm the change worked?</h3><p>Open a second session with your key and confirm it logs in, then check that sshd reports the setting as no. Keep the original session open until both checks pass.</p><p>Also confirm that a password attempt is refused, using a test account from a machine you control. If you run Match blocks, test from a connection that matches them, because a global check does not prove what those connections receive. Then read the file once more and search it for remaining PasswordAuthentication lines set to yes.</p><h3>Rollback</h3><p>If something breaks, use the session you kept open. Alpine’s undo is to delete the drop-in and restore the backup:</p><pre>rm /etc/ssh/sshd_config.d/00-no-passwords.conf<br>cp /etc/ssh/sshd_config.bak /etc/ssh/sshd_config<br>sshd -t &amp;&amp; rc-service sshd restart</pre><h3>Make it stick</h3><p>A manual change does not survive a rebuild, so put the setting in your configuration management or image build. Cloud images are the easy case to miss. Alpine notes that images configured through cloud-init write PasswordAuthentication no, while its tiny-cloud images do not. Check which one each fleet actually uses rather than assuming.</p><p>Then add a recurring check that reports any host whose effective value is yes. Run it across every platform, because <a href="https://www.anthonybahn.com/learn/why-musl-vs-glibc-security-differences-affect-patching/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=disable-ssh-password-authentication-safely">musl and glibc hosts carry different risks</a> but both benefit from having no password path on internet-facing SSH.</p><h3>This does not replace patching</h3><p>Disabling password logins closes the OpenSSH bypass, and Alpine says either the update or this change is enough. Treat it as a stopgap on affected hosts and still install the fixed package, then restart sshd. A later rebuild from an old image or a restored config file could bring password logins back, and an unpatched binary would then be exposed again.</p><p>Originally published at <a href="https://www.anthonybahn.com/learn/disable-ssh-password-authentication-safely/">https://www.anthonybahn.com/learn/disable-ssh-password-authentication-safely/</a></p><p><a href="https://ambwebmaster.gumroad.com/l/dnqemly?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=free-sheet">Free security one-pager (AMB Digital Group)</a></p><p>More at <a href="https://www.anthonybahn.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=disable-ssh-password-authentication-safely">anthonybahn.com</a> and <a href="https://ambdigitalgroup.substack.com/?utm_source=medium&amp;utm_medium=syndication&amp;utm_campaign=disable-ssh-password-authentication-safely">AMB Digital Group on Substack</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0c5a558ba8af" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>