<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Praesidia Blog</title>
    <link>https://praesidia.ai/blog</link>
    <atom:link href="https://praesidia.ai/feed.xml" rel="self" type="application/rss+xml"/>
    <description>See every agent you connect, control every action, prove what was decided — across the agents, models and tools you already run.</description>
    <language>en</language>
    <item>
      <title>Cisco AI Defense Alternatives for AI Agent Security</title>
      <link>https://praesidia.ai/blog/cisco-ai-defense-alternatives</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/cisco-ai-defense-alternatives</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Cisco AI Defense alternatives as of October 2026: what the product covers, where Cisco is strong, and which category fits when you need agent-level control.</description>
    </item>
    <item>
      <title>Credo AI and Holistic AI Alternatives for AI Governance</title>
      <link>https://praesidia.ai/blog/credo-ai-holistic-ai-alternatives-ai-governance</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/credo-ai-holistic-ai-alternatives-ai-governance</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>Credo AI and Holistic AI as of October 2026: how the two AI governance platforms compare, where each is strong, and when a runtime control plane fits too.</description>
    </item>
    <item>
      <title>Lasso Security Alternatives for AI Agent Security</title>
      <link>https://praesidia.ai/blog/lasso-security-alternatives</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/lasso-security-alternatives</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Lasso Security alternatives as of October 2026: discovery, red teaming and inline protection, its open-source MCP gateway, and how other categories compare.</description>
    </item>
    <item>
      <title>Microsoft Entra Agent ID Alternatives and Complements</title>
      <link>https://praesidia.ai/blog/microsoft-entra-agent-id-alternatives</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/microsoft-entra-agent-id-alternatives</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <category>Identity &amp; Access</category>
      <description>Microsoft Entra Agent ID as of October 2026: what agent identities, Conditional Access and ID Protection do for agents, and what a control plane adds.</description>
    </item>
    <item>
      <title>Pangea Alternatives for AI Guardrails and Agent Security</title>
      <link>https://praesidia.ai/blog/pangea-alternatives-ai-guardrails</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/pangea-alternatives-ai-guardrails</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Pangea alternatives as of October 2026: AI Guard and Pangea&apos;s security APIs, the CrowdStrike deal, and which category fits when you need more than guardrails.</description>
    </item>
    <item>
      <title>Prisma AIRS Alternatives for AI Agent Security</title>
      <link>https://praesidia.ai/blog/prisma-airs-alternatives-ai-security</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/prisma-airs-alternatives-ai-security</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Prisma AIRS alternatives as of October 2026: what Palo Alto Networks&apos; AI security platform covers, where it is strong, and when an agent control plane fits.</description>
    </item>
    <item>
      <title>AI Red Teaming Tools Compared: garak vs PyRIT vs promptfoo</title>
      <link>https://praesidia.ai/blog/ai-red-teaming-tools-compared-garak-pyrit-promptfoo</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/ai-red-teaming-tools-compared-garak-pyrit-promptfoo</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>garak scans models, PyRIT scripts multi-turn attacks, promptfoo tests applications. What each finds, what none of them see, and how to close the agent gap.</description>
    </item>
    <item>
      <title>Are Pickle Model Files Safe? Safetensors vs Pickle</title>
      <link>https://praesidia.ai/blog/are-pickle-model-files-safe-safetensors-explained</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/are-pickle-model-files-safe-safetensors-explained</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Pickle model files can run arbitrary code the moment they load. Why scanners miss payloads, how safetensors fixes it, and how to gate what agents load.</description>
    </item>
    <item>
      <title>What Is ASCII Smuggling? Invisible Unicode Injection</title>
      <link>https://praesidia.ai/blog/ascii-smuggling-unicode-tag-prompt-injection</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/ascii-smuggling-unicode-tag-prompt-injection</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>ASCII smuggling hides prompt-injection instructions and exfiltrated data in invisible Unicode tag characters. How it works and how to strip it both ways.</description>
    </item>
    <item>
      <title>Can LLMs Leak Their Training Data? Memorization Attacks</title>
      <link>https://praesidia.ai/blog/can-llms-leak-training-data-memorization</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/can-llms-leak-training-data-memorization</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>LLMs memorise parts of their training data and can be prompted to emit it verbatim. How extraction attacks work and the controls for fine-tunes and outputs.</description>
    </item>
    <item>
      <title>Can You Monitor Employee AI Usage Under GDPR?</title>
      <link>https://praesidia.ai/blog/can-you-monitor-employee-ai-usage-under-gdpr</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/can-you-monitor-employee-ai-usage-under-gdpr</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>Yes, if monitoring is proportionate, transparent and focused on what AI systems do, not on people. A GDPR checklist covering DPIA, works councils and logs.</description>
    </item>
    <item>
      <title>Computer-Use Agents: Security Risks of AI That Clicks</title>
      <link>https://praesidia.ai/blog/computer-use-agents-security-risks</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/computer-use-agents-security-risks</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Computer-use agents see the screen and drive mouse and keyboard with your session. Isolate them in a VM, allow-list actions, and gate irreversible steps.</description>
    </item>
    <item>
      <title>Does Fine-Tuning Break an LLM&apos;s Safety Guardrails?</title>
      <link>https://praesidia.ai/blog/does-fine-tuning-break-llm-safety-alignment</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/does-fine-tuning-break-llm-safety-alignment</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Often, yes. A few examples can erode refusals and narrow tasks can cause emergent misalignment. Re-test every fine-tune and keep guardrails outside the model.</description>
    </item>
    <item>
      <title>EchoLeak Explained: The Zero-Click M365 Copilot Exploit</title>
      <link>https://praesidia.ai/blog/echoleak-zero-click-copilot-prompt-injection-explained</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/echoleak-zero-click-copilot-prompt-injection-explained</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>EchoLeak (CVE-2025-32711) let one email make Microsoft 365 Copilot leak data with no user click. How the chain worked and which controls break it.</description>
    </item>
    <item>
      <title>Are Vector Embeddings Personal Data? Inversion Attacks</title>
      <link>https://praesidia.ai/blog/embedding-inversion-attacks-are-vector-embeddings-pii</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/embedding-inversion-attacks-are-vector-embeddings-pii</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>Embedding inversion can reconstruct text from vectors, so treat embeddings of personal data as personal data: erasure, residency and access duties apply.</description>
    </item>
    <item>
      <title>EU AI Act AI Literacy (Article 4): What Companies Must Do</title>
      <link>https://praesidia.ai/blog/eu-ai-act-ai-literacy-article-4-requirements</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/eu-ai-act-ai-literacy-article-4-requirements</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>Article 4 of the EU AI Act asks providers and deployers to support staff AI literacy. What it requires, how to evidence it, and how it ties to agent approvals.</description>
    </item>
    <item>
      <title>EU AI Act Article 73: Serious Incident Reporting Guide</title>
      <link>https://praesidia.ai/blog/eu-ai-act-serious-incident-reporting-article-73</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/eu-ai-act-serious-incident-reporting-article-73</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>Article 73 of the EU AI Act sets 2, 10 and 15-day clocks for reporting serious incidents with high-risk AI. What counts, who reports, and the evidence to keep.</description>
    </item>
    <item>
      <title>How to Govern ChatGPT and Claude Connectors</title>
      <link>https://praesidia.ai/blog/governing-chatgpt-claude-connectors-enterprise</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-chatgpt-claude-connectors-enterprise</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>MCP &amp; Integrations</category>
      <description>ChatGPT and Claude connectors link chat assistants to company data, often past IT review. Allow-list, scope and audit them like any other MCP server.</description>
    </item>
    <item>
      <title>How to Create an AI-BOM: CycloneDX vs SPDX</title>
      <link>https://praesidia.ai/blog/how-to-create-an-ai-bom-cyclonedx-spdx</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/how-to-create-an-ai-bom-cyclonedx-spdx</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>An AI bill of materials lists the models, datasets, tools and MCP servers an AI system depends on. How to build one, and CycloneDX ML-BOM vs SPDX 3.0.</description>
    </item>
    <item>
      <title>How Do You Measure Guardrail Accuracy and False Positives?</title>
      <link>https://praesidia.ai/blog/how-to-measure-guardrail-false-positives</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/how-to-measure-guardrail-false-positives</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Measure guardrails on block precision, recall, latency and bypass rate. Build a test set, replay real traffic in shadow mode, then set thresholds per policy.</description>
    </item>
    <item>
      <title>How to Set Up an AI Governance Committee</title>
      <link>https://praesidia.ai/blog/how-to-set-up-an-ai-governance-committee</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/how-to-set-up-an-ai-governance-committee</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>Set up an AI governance committee with a charter, RACI and meeting cadence. The committee decides policy; a control plane enforces it and feeds the board pack.</description>
    </item>
    <item>
      <title>How Do You Stop Employees Pasting Data into ChatGPT?</title>
      <link>https://praesidia.ai/blog/how-to-stop-employees-pasting-sensitive-data-into-chatgpt</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/how-to-stop-employees-pasting-sensitive-data-into-chatgpt</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Blocking ChatGPT drives use underground. Give staff a sanctioned path through a gateway with block and redact guardrails, then close the side doors.</description>
    </item>
    <item>
      <title>How to Threat Model an AI Agent (MAESTRO vs STRIDE)</title>
      <link>https://praesidia.ai/blog/how-to-threat-model-an-ai-agent-maestro</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/how-to-threat-model-an-ai-agent-maestro</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>A step-by-step AI agent threat model: map the layers with MAESTRO, enumerate threats per component with STRIDE, rank by blast radius and record the controls.</description>
    </item>
    <item>
      <title>Human-Agent Trust Exploitation (OWASP ASI09) Explained</title>
      <link>https://praesidia.ai/blog/human-agent-trust-exploitation-asi09</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/human-agent-trust-exploitation-asi09</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>OWASP ASI09 covers attacks that exploit human trust in agents. Approval fatigue turns HITL into theatre; risk-tiered approvals with evidence fix it.</description>
    </item>
    <item>
      <title>Is ChatGPT Safe for Company Data? Plans vs API</title>
      <link>https://praesidia.ai/blog/is-chatgpt-safe-for-company-data</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/is-chatgpt-safe-for-company-data</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Consumer ChatGPT, business plans and the API handle your data differently. Compare control, retention and training defaults, and when a governed API path wins.</description>
    </item>
    <item>
      <title>Is It Safe to Use DeepSeek and Open-Weight Models at Work?</title>
      <link>https://praesidia.ai/blog/is-it-safe-to-use-deepseek-at-work</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/is-it-safe-to-use-deepseek-at-work</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>DeepSeek&apos;s hosted app and its open weights carry different risks. Learn how to assess each, set a model allow-list, and govern use through a gateway.</description>
    </item>
    <item>
      <title>ISO 27001 vs ISO 42001: Do You Need Both?</title>
      <link>https://praesidia.ai/blog/iso-27001-vs-iso-42001</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/iso-27001-vs-iso-42001</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>ISO 27001 secures information; ISO 42001 governs AI systems. How the two management systems overlap, where 42001 adds AI controls, and when you need both.</description>
    </item>
    <item>
      <title>Can an LLM Be Backdoored? Sleeper Agents and Data Poisoning</title>
      <link>https://praesidia.ai/blog/llm-backdoors-sleeper-agents-data-poisoning</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/llm-backdoors-sleeper-agents-data-poisoning</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Yes. Poisoned training data can plant hidden triggers that survive safety training. Why runtime monitoring and drift detection are the practical backstop.</description>
    </item>
    <item>
      <title>MCP Sampling and Elicitation: Security Risks</title>
      <link>https://praesidia.ai/blog/mcp-sampling-elicitation-security-risks</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/mcp-sampling-elicitation-security-risks</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>MCP &amp; Integrations</category>
      <description>MCP sampling lets servers call your LLM; elicitation lets them prompt your users. Both open injection and phishing channels. Here are the risks and controls.</description>
    </item>
    <item>
      <title>Why MCP Token Passthrough Is Forbidden</title>
      <link>https://praesidia.ai/blog/mcp-token-passthrough-anti-pattern</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/mcp-token-passthrough-anti-pattern</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>MCP &amp; Integrations</category>
      <description>MCP servers must not accept or forward tokens issued for another resource. Why passthrough breaks audit and trust, and audience-bound patterns to use instead.</description>
    </item>
    <item>
      <title>How to Verify AI Model Provenance with Sigstore Signing</title>
      <link>https://praesidia.ai/blog/model-signing-provenance-openssf-sigstore</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/model-signing-provenance-openssf-sigstore</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Verify an AI model&apos;s provenance by checking a signature over every model file against a trusted signer identity, with OpenSSF model signing and Sigstore.</description>
    </item>
    <item>
      <title>Multimodal Prompt Injection: Images, PDFs and Audio</title>
      <link>https://praesidia.ai/blog/multimodal-prompt-injection-images-audio-documents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/multimodal-prompt-injection-images-audio-documents</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Multimodal prompt injection hides instructions in images, PDFs and audio that text filters never see. Learn where payloads hide and which controls hold.</description>
    </item>
    <item>
      <title>NIST Cyber AI Profile Explained: CSF 2.0 for AI Systems</title>
      <link>https://praesidia.ai/blog/nist-cyber-ai-profile-csf-2-explained</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/nist-cyber-ai-profile-csf-2-explained</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>NIST IR 8596, the Cyber AI Profile, maps CSF 2.0 outcomes to AI across Secure, Defend and Thwart. What it asks for and how to map it to agent controls.</description>
    </item>
    <item>
      <title>How Do You Enforce Document Permissions in RAG?</title>
      <link>https://praesidia.ai/blog/permission-aware-rag-document-level-access-control</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/permission-aware-rag-document-level-access-control</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Permission-aware RAG filters retrieval by the caller&apos;s entitlements at query time and logs which chunks reached whom. How to build it for humans and agents.</description>
    </item>
    <item>
      <title>How to Secure AI SOC Agents Against Prompt Injection</title>
      <link>https://praesidia.ai/blog/securing-ai-soc-agents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/securing-ai-soc-agents</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>AI SOC agents read attacker-controlled logs and alerts. Scope their response actions, gate containment behind humans, and audit every autonomous verdict.</description>
    </item>
    <item>
      <title>How to Secure a Self-Hosted LLM: Ollama and vLLM Checklist</title>
      <link>https://praesidia.ai/blog/self-hosted-llm-security-ollama-vllm-hardening</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/self-hosted-llm-security-ollama-vllm-hardening</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Self-hosted LLM servers often ship without auth. A hardening checklist for Ollama and vLLM: network binding, auth, isolation, model provenance and audit.</description>
    </item>
    <item>
      <title>What Is Slopsquatting? AI Package Hallucination Attacks</title>
      <link>https://praesidia.ai/blog/slopsquatting-ai-package-hallucination-attacks</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/slopsquatting-ai-package-hallucination-attacks</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Slopsquatting is registering package names that AI coding tools hallucinate, so agents install malware. How it works and how to block it with allow-lists.</description>
    </item>
    <item>
      <title>Who Is Liable When an AI Agent Makes a Mistake?</title>
      <link>https://praesidia.ai/blog/who-is-liable-when-an-ai-agent-makes-a-mistake</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/who-is-liable-when-an-ai-agent-makes-a-mistake</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>When an AI agent errs, the company deploying it is usually liable first. How contract, negligence and the new EU Product Liability Directive split the risk.</description>
    </item>
    <item>
      <title>What Is Zero Data Retention (ZDR) for LLM Providers?</title>
      <link>https://praesidia.ai/blog/zero-data-retention-llm-providers-explained</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/zero-data-retention-llm-providers-explained</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>Zero data retention means a provider stores no prompts or outputs after the request. Learn how ZDR works, where it stops, and how to enforce it.</description>
    </item>
    <item>
      <title>Verify Signed AI Agent Audit Logs: A Runnable Example</title>
      <link>https://praesidia.ai/blog/signed-ai-agent-audit-log-verification-example</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/signed-ai-agent-audit-log-verification-example</guid>
      <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance &amp; Compliance</category>
      <description>Run a synthetic signed-audit lab to detect changed records, broken chains, and missing tails with a trusted checkpoint. Learn what verification cannot prove.</description>
    </item>
    <item>
      <title>Fix MCP OAuth 401 and 403 Errors</title>
      <link>https://praesidia.ai/blog/mcp-oauth-troubleshooting-401-403</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/mcp-oauth-troubleshooting-401-403</guid>
      <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>
      <category>MCP</category>
      <description>Diagnose MCP OAuth failures with a local lab: missing or expired tokens, wrong audiences, insufficient scopes, discovery problems, and issuer mismatches.</description>
    </item>
    <item>
      <title>EU AI Act Article 12 Logging for AI Agents</title>
      <link>https://praesidia.ai/blog/eu-ai-act-article-12-logging-ai-agents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/eu-ai-act-article-12-logging-ai-agents</guid>
      <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>
      <category>AI Governance &amp; Compliance</category>
      <description>What EU AI Act Article 12 record-keeping means for AI agents that call tools: the events to log, who keeps the logs under Articles 19 and 26(6), and how long.</description>
    </item>
    <item>
      <title>The EU Cyber Resilience Act and AI Agent Products</title>
      <link>https://praesidia.ai/blog/eu-cyber-resilience-act-ai-agent-products</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/eu-cyber-resilience-act-ai-agent-products</guid>
      <pubDate>Sun, 06 Sep 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>How the EU Cyber Resilience Act&apos;s security-by-design, vulnerability handling, and reporting duties apply to AI agent software sold into the EU market.</description>
    </item>
    <item>
      <title>Governing Workday Agents Over HR Data</title>
      <link>https://praesidia.ai/blog/governing-workday-illuminate-agents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-workday-illuminate-agents</guid>
      <pubDate>Sat, 05 Sep 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>Workday&apos;s domain security model scopes agent access to HR data tightly, but integrations to external HRIS, payroll, and ATS systems fall outside it.</description>
    </item>
    <item>
      <title>Governing Snowflake Cortex Agents</title>
      <link>https://praesidia.ai/blog/governing-snowflake-cortex-agents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-snowflake-cortex-agents</guid>
      <pubDate>Fri, 04 Sep 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>Snowflake&apos;s RBAC and row-level policies govern Cortex Agents tightly inside the platform, but external functions and exports fall outside that perimeter.</description>
    </item>
    <item>
      <title>Governing SAP Joule Agents in Regulated Estates</title>
      <link>https://praesidia.ai/blog/governing-sap-joule-agents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-sap-joule-agents</guid>
      <pubDate>Thu, 03 Sep 2026 00:00:00 GMT</pubDate>
      <category>Compliance</category>
      <description>Joule inherits SAP&apos;s authorization objects in-landscape, but regulated estates need an audit trail spanning any system a Joule agent reaches beyond SAP.</description>
    </item>
    <item>
      <title>Governing Salesforce Agentforce Deployments</title>
      <link>https://praesidia.ai/blog/governing-salesforce-agentforce-deployments</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-salesforce-agentforce-deployments</guid>
      <pubDate>Wed, 02 Sep 2026 00:00:00 GMT</pubDate>
      <category>Agent Security</category>
      <description>Agentforce agents run as a dedicated Salesforce agent user, not the invoking human, so what that user&apos;s permission sets grant is the first thing to check.</description>
    </item>
    <item>
      <title>Governing Microsoft 365 Copilot Agents</title>
      <link>https://praesidia.ai/blog/governing-microsoft-365-copilot-agents</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-microsoft-365-copilot-agents</guid>
      <pubDate>Tue, 01 Sep 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>Copilot agents respect Microsoft Graph permissions by default, but citizen-developer publishing in Copilot Studio outpaces central inventory and review.</description>
    </item>
    <item>
      <title>Governing Google Vertex AI Agent Engine Deployments</title>
      <link>https://praesidia.ai/blog/governing-google-vertex-ai-agent-engine</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-google-vertex-ai-agent-engine</guid>
      <pubDate>Mon, 31 Aug 2026 00:00:00 GMT</pubDate>
      <category>Identity &amp; Access</category>
      <description>What Vertex AI Agent Engine&apos;s IAM-based identity and logging cover for agents on Google Cloud, and where multi-cloud and MCP-connected tools fall outside it.</description>
    </item>
    <item>
      <title>Governing Agents on Databricks</title>
      <link>https://praesidia.ai/blog/governing-databricks-agent-framework-mosaic</link>
      <guid isPermaLink="true">https://praesidia.ai/blog/governing-databricks-agent-framework-mosaic</guid>
      <pubDate>Sun, 30 Aug 2026 00:00:00 GMT</pubDate>
      <category>AI Governance</category>
      <description>Unity Catalog governs data and tool access tightly inside Databricks, but agents that reach outside it or span multi-cloud workspaces need separate controls.</description>
    </item>
  </channel>
</rss>
