<?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 Ruchi Mishra Gautam on Medium]]></title>
        <description><![CDATA[Stories by Ruchi Mishra Gautam on Medium]]></description>
        <link>https://medium.com/@ruchimishragautam?source=rss-650b8509ffdd------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*_LuBpoCjo5QOYv_B.</url>
            <title>Stories by Ruchi Mishra Gautam on Medium</title>
            <link>https://medium.com/@ruchimishragautam?source=rss-650b8509ffdd------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 15:08:11 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ruchimishragautam/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[Stopping Isn’t Rollback: Designing Kill Switches for Agentic AI]]></title>
            <link>https://medium.com/@ruchimishragautam/stopping-isnt-rollback-designing-kill-switches-for-agentic-ai-4eeaa5641e10?source=rss-650b8509ffdd------2</link>
            <guid isPermaLink="false">https://medium.com/p/4eeaa5641e10</guid>
            <category><![CDATA[trust-and-safety]]></category>
            <category><![CDATA[eu-ai-act-compliance]]></category>
            <category><![CDATA[ai-kill-switch]]></category>
            <category><![CDATA[conversational-ai]]></category>
            <dc:creator><![CDATA[Ruchi Mishra Gautam]]></dc:creator>
            <pubDate>Thu, 01 Oct 2026 03:34:37 GMT</pubDate>
            <atom:updated>2026-10-01T03:34:37.624Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s 4:02. A customer-support assistant is triaging the overnight backlog. It reads each ticket, decides whether the complaint qualifies under the refund policy, writes a reply in the company’s voice, and sends it. There are forty-one conversations, a ticket per conversation to handle.</p><p>Each ticket is a new session, the session making its own model calls and its own tool calls. The assistant lookups against the order system, writes to the ticket, issues credit in case of a refund, and sends email. Eleven sessions have handed slow cases off to a second assistant that waits on a billing lookup. Nine replies are sitting in the outbound mail queue on a thirty-second flush. The mail is written using the service account that was issued a token at 3:58 and is good for fifteen minutes.</p><p>At 4:06 a supervisor reads one of the replies and goes cold. The Agent has been telling customers they’re ineligible for refunds they’re entitled to (policy information update made today, not available with the agent yet).</p><p>They flip the kill switch.</p><p>Thankfully, the gateway stops accepting new conversations. Active sessions drop to zero. The dashboard goes green. Someone types “killed it” in the incident channel, everyone stops working on the incident as they consider it addressed!!</p><p>Is it addressed really?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Wm8CnkkaPVIB0a-FfAifew.png" /><figcaption>Case in point, lack of Trust Control Zone</figcaption></figure><p>Most teams building Agentic AI have a kill switch in the sense that there is a flag that turns things off. Have we thought about what state the system is in once that flag is set — which is the harder problem.</p><p>Why are we talking about it?</p><p>It’s the most frequently penned down topic in the AI news right now. The reasons are not just due to this topic being proposed as The AI KillSwitch Act in the legislature, or The EU AI Act wants the High Risk AI systems handled.Ref. <a href="https://artificialintelligenceact.eu/article/14/#p-article-14-4.e">Article 14–4(e)</a> . The incidents of AI going rouge has raised an alarm to an extent that the officials fear massive level data may be wiped out by AI Agents.</p><p>My article above gives a case in point of what lack of trust control zone handling Kill Switch within an Org or a team can look like. Put on your Kill Switch thinking hat on how the Trusted Zone looks like for your Agentic System.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4eeaa5641e10" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Services and digital certificates]]></title>
            <link>https://medium.com/@ruchimishragautam/ai-services-and-digital-certificates-d14624df8b99?source=rss-650b8509ffdd------2</link>
            <guid isPermaLink="false">https://medium.com/p/d14624df8b99</guid>
            <category><![CDATA[pki]]></category>
            <category><![CDATA[ai-native]]></category>
            <category><![CDATA[digital-certificates]]></category>
            <category><![CDATA[ai-workflow-automation]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Ruchi Mishra Gautam]]></dc:creator>
            <pubDate>Sat, 07 Feb 2026 16:55:35 GMT</pubDate>
            <atom:updated>2026-02-07T16:55:35.405Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Hidden Infrastructure of AI Agents: How Digital Certificates Will Evolve in the AI-Agentic workflows</h3><p>We are no longer in the era of AI revolution happening in chatbots and automation workflows, the enterprises have moved to Agentic workflows. Behind every autonomous agent making decisions, processing data, and executing tasks lies a complex web of trust. <strong>How do we prove an AI agent is who it claims to be?</strong></p><p>As enterprises embrace AI agents to handle everything from customer service to supply chain optimization to autonomous offices, we’re witnessing a fundamental shift in how digital identity and security must operate. The AI-native platforms enabling this transformation — spanning automation orchestration, AIaaS, database management, ML operations, and security layers — all face a common challenge: establishing and maintaining trust in a world where software entities outnumber human users by orders of magnitude.</p><h3>The Certificate Crisis, it is real</h3><p>Traditional digital certificate lifecycle management was designed for a relatively static world. You issue certificates to servers, employees get credentials, applications are authenticated, and certificates are renewed on predictable cycles — often manually, sometimes taking months to deploy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W8WWGA1HMr7YXoWdtBQ1SQ.png" /></figure><p>We moved from cumbersome, time taking traditional digital certificate management systems to cloud based microservices systems. The frequency of issuance jumped from monthly to daily, increasing scale by 100X.</p><p>Now consider an AI agent ecosystem: hundreds or thousands of autonomous agents spinning up and down dynamically, making API calls across multiple services, accessing sensitive data, and acting on behalf of your organization. Each agent needs its own identity. Each interaction requires authentication. Each service-to-service communication demands encryption and verification.</p><p>The math is staggering. Where an enterprise might manage tens of thousands of certificates today, an AI-native organization could be managing millions within months.</p><h3>Certificates are handled by AI-Native Platforms, is autonomous generation the way to go?</h3><p>The enterprises on microservices cloud architecture are automating certificate issuances, managing expiration and renewals in seconds. The ephemeral nature and policy driven approach helps organizations speed up the certificates delivery from monthly to daily to seconds. With Agentic AI certificate deliveries SLAs are in seconds. Credential generation happens just in time, tied to an agent lifecycle events. Policy driven certificate templates will be used to apply right permissions based on agent function.</p><p>Is there a middle approach to automation or JUST GO autonomous?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IZytITwEELdpcP4bVcFjUQ.png" /></figure><p>Monolithic architecture is replaced by Layers of specialized AI. Each layer requires its own security posture. Adapting Hierarchical PKI infrastructure is the choice most platforms are making.</p><p><strong>Application Layer</strong> (Where agents live)</p><ul><li>Agent-to-agent authentication using mutual TLS</li><li>Service mesh certificates for microservices communication</li><li>API gateway certificates for external integrations</li></ul><p><strong>AIaaS Layer</strong> (Model serving and inference)</p><ul><li>Model endpoint certificates ensuring you’re calling legitimate AI services</li><li>Credentials for model versioning and deployment pipelines</li><li>Secure channels for sensitive inference requests (PII, financial data)</li></ul><p><strong>Data Layer</strong> (Databases, data lakes, vector stores)</p><ul><li>Database connection certificates that rotate automatically</li><li>Encryption certificates for data at rest</li><li>Access credentials tied to specific agent roles and data scopes</li></ul><p><strong>Infrastructure Layer</strong> (Compute, storage, networking)</p><ul><li>Node certificates for distributed AI training</li><li>Storage encryption keys managed through certificate-based key hierarchies</li><li>Network segmentation enforced through certificate-based policies</li></ul><h3>The Platform Integration Challenge</h3><p>The most sophisticated certificate lifecycle systems emerging today don’t exist as standalone tools, they’re woven into the fabric of AI platforms:</p><ul><li><strong>Kubernetes operators</strong> that automatically provision certificates for AI workloads</li><li><strong>Service mesh integration</strong> (Istio, Linkerd) that handles agent-to-agent mTLS transparently</li><li><strong>Secret management systems</strong> (HashiCorp Vault, AWS Secrets Manager) that store and rotate agent credentials</li><li><strong>Policy engines</strong> (OPA, Kyverno) that enforce certificate requirements before agents can deploy</li><li><strong>Observability stacks</strong> that correlate agent behavior with credential usage</li></ul><p>The best platforms present a unified control plane where security teams can define certificate policies once and have them automatically enforced across all layers.</p><p><strong>As per Gemini, Top Digital PKI Platforms for AI-Native Systems:</strong></p><ul><li><strong>DigiCert ONE (and DigiCert + QuoVadis)</strong>: Often cited as a leading solution for AI-powered, automated PKI, focusing on high-volume issuance, quantum-safe cryptography, and unified management of digital trust for AI content and infrastructure.</li><li><strong>Sectigo:</strong> Provides cloud-native, automated Certificate Lifecycle Management (CLM) for securing both machine and human identities at scale.</li><li><strong>Keyfactor:</strong> Focuses on integrating AI to optimize PKI management, increasing efficiency and responsiveness in high-stakes, automated environments.</li><li><strong>SecureW2 JoinNow Platform:</strong> Offers Managed PKI services with cloud-based RADIUS for robust user and device authentication.</li></ul><h3>Amplifying the Micro Services Architecture</h3><p>The AI Agents are going to be integrated into your platforms sooner or later, amplifying the Micro Services within your platform architecture. Hence, identifying the opportunities of automating operations is the best possible approach. Maybe your third party MDR service providers have started incorporating the use cases of Agentic AI led certificate lifecycle management. Even if you automate the processes, there will be manual steps with human-in-the-loop operations, static policies that don’t adapt, Reactive incident response, manual compliance processes and fragmented multi-cloud management. The CTO/COO office will announce Native AI integration if they haven’t yet. Set yourself up with a risk analysis and how to set up a zero trust infrastructure using Agents.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d14624df8b99" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is the Voice Assistant privacy ready?]]></title>
            <link>https://medium.com/@ruchimishragautam/is-the-voice-assistant-privacy-ready-188c8135a729?source=rss-650b8509ffdd------2</link>
            <guid isPermaLink="false">https://medium.com/p/188c8135a729</guid>
            <category><![CDATA[data-compliance]]></category>
            <category><![CDATA[data-act]]></category>
            <category><![CDATA[data-privacy]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <dc:creator><![CDATA[Ruchi Mishra Gautam]]></dc:creator>
            <pubDate>Thu, 19 Sep 2024 00:38:56 GMT</pubDate>
            <atom:updated>2024-09-19T00:38:56.795Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aUiOvo7kay03VvZY" /><figcaption>Photo by <a href="https://unsplash.com/@qbuffing?utm_source=medium&amp;utm_medium=referral">Quinn Buffing</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Mimics of famous personalities entertain us. They are artists, creators of comedy. Imagine your Voice Assistant responds to you in your Mom’s voice!! Voice can be used/misused as impersonation tools. The big tech companies are relying on speech recognition technologies to sell convenience to their users. In turn, by default, they are capturing our voices too. Do they really need to capture EVERYTHING that gets spoken into the devices( Alexa, Siri, Google etc.) (including your vocal cords synchronization - with Voice Recognition technology the devices have been programmed to capture the voice biometric details) !</p><p>Speech recognition, on the other hand is the process of listening to the speech ( series of texts), converting the speech to text format and using the information to process the response. User might provide personal data information to Voice Assistant for processing the response. Should this grant rights of data be stored by the Propreitory owner of the Speech recognition technology or their technology partners owning the device ? Is the company just providing a service or ACTUALLY storing data at the cost of user’s trust?</p><p>Some of the best practices in place as of today for development and maintenence of Speech Recognitions products and solutions to make sure you are complying with latest laws and reducing your Risk.</p><ul><li>Responsible AI development for a safe and non-toxic use</li><li>Anonymization of data is important to safe guard privacy of personal data</li><li>Need to know data only should be kept, there needs to be a business justification for storing the personal data</li><li>Secure deletion of the data that is not required for generating the response</li></ul><p>Here are some policies to be aware of.</p><ol><li><a href="https://gdpr.eu/article-13-personal-data-collected/">Article 13 of GDPR Policy</a> states details of obligations on developers in the event of collecting personal data or processing it. This policy came into effect since May 2018. It is applicable to all data subjects based in Europe. Both Voice data being a biometric data, and Speech text information comprising of personal data information are to comply with GDPR.</li><li><a href="https://digital-strategy.ec.europa.eu/en/policies/data-act#:~:text=The%20Data%20Act%20gives%20individuals,this%20data%20with%20third%20parties.">Data Act Nov 2023</a> lays down laws applicable in EU on rights applicable to businesses and subjects of data. This compliments some of the policy from GDPR, whilst clarifying the data usage policies for connected devices. Since Voice Assistants and Connected Devices are 2 sides of the same coin, it is inevitable to not develop your Voice Assistant product or service without complying to this Act. It is going to be effective September 2025.</li><li>Security Operations Center Certification</li><li><a href="https://www.gsa.gov/about-us?_gl=1*muh4xo*_ga*MTU3MTMyMDYwMi4xNzI2NTU0NDg1*_ga_HBYXWFP794*MTcyNjU1NDQ4NC4xLjEuMTcyNjU1NDU2Ni4wLjAuMA..">GSA</a> and The Federal Risk and Authorization Management Program (FedRAMP®) Compliance</li><li>HIPAA compliance — yes, you do need to secure health information being shared with the voice assistants.</li><li><a href="https://www.pcisecuritystandards.org/document_library/?category=pcidss&amp;document=pci_dss">PCI DSS</a> compliance requirements — Whether or not user has ever shared their credit card information, it is important to protect your customers from a credit card fraud.</li></ol><p>Why should you care, see the following snippets of consequences of non-compliance:</p><ul><li>As per a research paper (<em>Understanding GDPR Non-Compliance in Privacy Policies of Alexa Skills in European Marketplaces </em>), <a href="https://dl.acm.org/doi/pdf/10.1145/3589334.3645409">Policy compliance check in Amazon Alexa was not met in 95% cases</a>.</li><li>See Google Voice Assistant Privacy report <a href="https://privacy.commonsense.org/evaluation/Google-Assistant">here</a> from a 3rd party company, issuing warning as per their parameters(worth checking to weigh your risks).</li><li>A case of <a href="https://nyidanmark.dk/en-GB/News-Front-Page/2023/03/Databrud-i-SIRI">Siri non-compliance</a>, an employee accessing personal data, was terminated from the job.</li></ul><p>As the connected devices surpass the human population on earth, Voice Assistant product and services are ubiquitous. GenAI models will add to the proliferation of use cases that will require Compliance implementation based on the geographies the services are launched in. You don’t want to be part of the list of penalty paid like <a href="https://www.csoonline.com/article/567531/the-biggest-data-breach-fines-penalties-and-settlements-so-far.html">these</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=188c8135a729" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>