<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>AI Agent Security &amp; Governance on Kevin Keller</title><link>https://kevinkeller.org/topics/ai-agent-security/</link><description>Recent content in AI Agent Security &amp; Governance on Kevin Keller</description><generator>Hugo -- gohugo.io</generator><language>en</language><managingEditor>kellerkev@gmail.com (Kevin Keller)</managingEditor><webMaster>kellerkev@gmail.com (Kevin Keller)</webMaster><copyright>© 2026 Kevin Keller</copyright><lastBuildDate>Fri, 13 Nov 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://kevinkeller.org/topics/ai-agent-security/index.xml" rel="self" type="application/rss+xml"/><item><title>One Policy Engine for Every Agent: Snowflake as the Enforcement Point for Cedar and AuthZEN</title><link>https://kevinkeller.org/posts/snowflake-agent-authz-external-pdp-cedar-authzen/</link><pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/snowflake-agent-authz-external-pdp-cedar-authzen/</guid><description>Many security teams want one central policy decision point for all AI agents, with Cedar policies behind the OpenID AuthZEN API. A working demo with real Entra ID identities shows how Snowflake enforces those decisions per column, per row and per session, without becoming a second policy engine, and why that holds even when the gateway is bypassed or the PDP goes down.</description></item><item><title>Keep Shipping: How to Say Yes to AI Agents Without Losing Control</title><link>https://kevinkeller.org/posts/keep-shipping-ai-agent-security-snowflake/</link><pubDate>Thu, 12 Nov 2026 12:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/keep-shipping-ai-agent-security-snowflake/</guid><description>AI agent projects don&amp;rsquo;t die on cost. They die on trust. Blocking agents only works if you can identify them, and point tools can&amp;rsquo;t secure agents. Only a platform can, at the data layer, the network layer, or both. The five controls every platform must provide, and how to get agents from pilot to production.</description></item><item><title>Snowflake Cortex AI -- Governance, Audit &amp; Inference Region Guide</title><link>https://kevinkeller.org/posts/snowflake-cortex-ai-governance-audit-inference-guide/</link><pubDate>Fri, 03 Apr 2026 07:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/snowflake-cortex-ai-governance-audit-inference-guide/</guid><description>Complete guide to Snowflake Cortex AI governance &amp;ndash; audit logging across all usage views, inference region controls for EU data residency, model-level RBAC with CORTEX_MODELS_ALLOWLIST, Cortex Analyst fallback chains, AI observability, REST API audit coverage, and the full model catalog with hosting infrastructure details.</description></item><item><title>Beyond RBAC — Why AI Agents Need Purpose-Aware Access Control</title><link>https://kevinkeller.org/posts/beyond-rbac-purpose-aware-access-control-ai-agents/</link><pubDate>Tue, 31 Mar 2026 12:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/beyond-rbac-purpose-aware-access-control-ai-agents/</guid><description>RBAC tells you if a role can access a table. But can this agent invoke this tool on this data for this purpose? The industry is building the pieces — Cedar, Proofpoint, Cisco, Immuta — but the unified policy engine that evaluates all attributes across all layers doesn&amp;rsquo;t exist yet.</description></item><item><title>Federal Zero Trust Architecture Blueprint — Snowflake as the AI + Data Control Plane</title><link>https://kevinkeller.org/posts/snowflake-federal-zero-trust-architecture/</link><pubDate>Fri, 27 Mar 2026 10:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/snowflake-federal-zero-trust-architecture/</guid><description>Snowflake provides a unified Zero Trust control plane for both data and AI inference. This blueprint maps Snowflake and Cortex to NIST 800-207, CISA&amp;rsquo;s Zero Trust Maturity Model, and OMB M-22-09 — one governance layer, one policy stack, no shadow AI.</description></item><item><title>SMCP: Adding Security to the Model Context Protocol</title><link>https://kevinkeller.org/posts/smcp-secure-model-context-protocol/</link><pubDate>Thu, 26 Mar 2026 08:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/smcp-secure-model-context-protocol/</guid><description>A proof-of-concept that layers authentication, encryption, and multi-agent coordination onto MCP — with four security tiers from API keys to enterprise OAuth2 with audit trails.</description></item><item><title>Containing AI Agents: A Defense-in-Depth Architecture for Production</title><link>https://kevinkeller.org/posts/containing-ai-agents-production/</link><pubDate>Thu, 19 Mar 2026 12:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/containing-ai-agents-production/</guid><description>Generic agents can do anything — which is exactly the problem. Custom agents in containers behind an inference proxy create a production architecture where security is structural, not aspirational.</description></item><item><title>Building the Agentic Enterprise: From Protocol Hype to Production-Grade AI Agents</title><link>https://kevinkeller.org/posts/building-agentic-enterprise-custom-agents-production/</link><pubDate>Thu, 19 Mar 2026 00:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/building-agentic-enterprise-custom-agents-production/</guid><description>The agentic enterprise is real. But getting there requires a shift from generic protocol-based integrations to purpose-built agents with native tooling, embedded security, and domain-specific intelligence.</description></item><item><title>Cortex Proxy: Route Any AI Coding Agent Through Snowflake Cortex, OpenAI, Anthropic, or Ollama — With Prompt Policy Enforcement</title><link>https://kevinkeller.org/posts/snowflake-cortex-proxy-ai-coding-agents/</link><pubDate>Wed, 18 Mar 2026 00:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/snowflake-cortex-proxy-ai-coding-agents/</guid><description>A high-performance Rust proxy that lets Claude Code, Continue.dev, ZeroClaw, Mistral Vibe — or any AI coding agent — use Snowflake Cortex, OpenAI, Anthropic, or Ollama as their backend. Optional prompt policy enforcement included.</description></item><item><title>Governing AI Inference in the Data Cloud: A Security Architecture Guide for EMEA Enterprises</title><link>https://kevinkeller.org/posts/governing-ai-inference-data-cloud/</link><pubDate>Tue, 17 Mar 2026 12:00:00 +0000</pubDate><author>kellerkev@gmail.com (Kevin Keller)</author><guid>https://kevinkeller.org/posts/governing-ai-inference-data-cloud/</guid><description>A defense-in-depth security architecture for AI agents and inference workloads — covering network isolation, identity propagation, authorization, data protection, and auditing under the EU AI Act, DORA, and NIS2.</description></item></channel></rss>