<?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 Mukund Mohan on Medium]]></title>
        <description><![CDATA[Stories by Mukund Mohan on Medium]]></description>
        <link>https://medium.com/@mukund?source=rss-7a697b7ed7e1------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*WQshbDpIELgsS9vu.JPG</url>
            <title>Stories by Mukund Mohan on Medium</title>
            <link>https://medium.com/@mukund?source=rss-7a697b7ed7e1------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 16:58:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mukund/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[We Analyzed 53 AI Policies. Here’s Where They All Fail.]]></title>
            <link>https://medium.com/@mukund/we-analyzed-53-ai-policies-heres-where-they-all-fail-3f193f615d7a?source=rss-7a697b7ed7e1------2</link>
            <guid isPermaLink="false">https://medium.com/p/3f193f615d7a</guid>
            <category><![CDATA[shadow-ai]]></category>
            <category><![CDATA[ai-tools]]></category>
            <category><![CDATA[shadow-ai-detection]]></category>
            <dc:creator><![CDATA[Mukund Mohan]]></dc:creator>
            <pubDate>Thu, 09 Apr 2026 04:34:55 GMT</pubDate>
            <atom:updated>2026-04-10T15:06:10.141Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Every company is writing an AI policy right now. Most were written with ChatGPT. Nearly all of them have the same blind spots.</em></p><p>Over the last 5 months, we helped 121 companies prepare for their SOC2 renewal and validation. Companies using the top SOC2 tools such as <a href="https://www.vanta.com/">Vanta</a>, <a href="https://drata.com/">Drata</a> and <a href="https://secureframe.com/">Secureframe</a>.</p><p>We <a href="https://benchmark.peridot.company/">Analyzed 53</a> actual AI Policies submitted to auditors. Here’s Where They All Fail.</p><p>Every company is writing an AI policy right now. Most of them were written with ChatGPT. The irony is not lost on us.</p><h3>The Actual Prompts They Use</h3><p>These are not hypothetical. This is <strong>exact behavior pattern</strong>.</p><h3>1. The Generic Policy Prompt</h3><blockquote><em>“Write an internal AI usage policy for a SaaS company handling customer data”</em></blockquote><p>OR</p><blockquote><em>“Create a policy for employees using ChatGPT and AI tools securely”</em></blockquote><p><strong>Output they want: </strong>Clean, Formal &amp; Sounds like they have control</p><p><strong>Reality: </strong>Zero enforcement mechanism</p><h3>2. The Compliance-Framed Prompt</h3><blockquote><em>“What controls should we have for AI usage to pass SOC 2?”</em></blockquote><p>OR</p><blockquote><em>“How do companies govern employee use of LLMs with sensitive data?”</em></blockquote><p><strong>Why this matters:</strong></p><ul><li>They’re reverse-engineering compliance language</li><li>Not solving the actual problem</li></ul><h3>3. The Risk Checklist Prompt</h3><blockquote><em>“What are the risks of employees using ChatGPT internally?”</em></blockquote><p>OR</p><blockquote><em>“How do we prevent data leakage from AI tools?”</em></blockquote><p><strong>What they get:</strong></p><ul><li>Data leakage</li><li>IP exposure</li><li>Hallucinations</li><li>Model training concerns</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EMsW4Q41xZBjoWjslU3UNg.png" /></figure><p><strong>What’s missing:</strong></p><ul><li>Visibility into <strong>internal AI-built apps/workflows</strong></li></ul><p>We’ve reviewed over 53 enterprise AI policies across industries. They’re thoughtful documents. They cover the obvious ground. And they all share the same catastrophic assumption.</p><p>What the policies cover<br>✓ Acceptable use guidelines<br>✓ Data handling and classification rules<br>✓ Approved and restricted tool lists</p><p><strong>Missing</strong>:</p><p><strong>“What AI apps actually exist inside our company?” — None of them answer this.</strong></p><p>This is not a minor gap. It’s the gap that makes everything else decorative.</p><p>The 5 Failure Points</p><p>1. “Approved Tools Only”<br>Assumes: you already know what tools are being used.<br>Reality: you don’t. Shadow AI is invisible by design.</p><p>2. “No Sensitive Data in AI Tools”<br>Assumes: employees will self-police their behavior.<br>Reality: no enforcement mechanism, no detection, no signal.</p><p>3. “Register Internal AI Apps”<br>Assumes: only developers are building AI-powered tools.<br>Reality: ops, marketing, and finance are building them too — in Zapier, Notion, and Make.</p><p>4. “Audit &amp; Monitoring”<br>Assumes: logs exist somewhere to review.<br>Reality: AI workflows live outside your observability stack entirely.</p><p>5. “Employee Responsibility”<br>Translation: “we hope people behave.”<br>Reality: hope is not a control framework.</p><p>— -</p><p>AI policies are written as if visibility already exists.<br>In most companies, it doesn’t.</p><p>— -</p><p>If you can’t answer the question “what AI apps exist in our org right now?” — your policy isn’t control. It’s documentation. And documentation doesn’t stop a breach.</p><p>The companies that get this right aren’t writing better policies. They’re building the visibility layer first, then letting the policy reflect what’s actually there.</p><p>We built a <a href="https://peridot.company/security/ai-risk-audit">policy stress test</a> to show exactly where yours breaks.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3f193f615d7a" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>