AI agent security, identity, governance, cost, and the engineering behind the control plane.
The EU AI Act Digital Omnibus entered into force 27 July 2026, deferring Annex III high-risk rules to Dec 2027 and Annex I to Aug 2028. Art. 49/50 unchanged.
What AI agent runtime security means: in-band enforcement of identity, authorization, guardrails, resource caps, and kill switches while agents execute.
Spend caps and request throttling are different levers for controlling runaway AI agents. Learn when each applies, how they compose, and why you need both.
How Praesidia continuously red-teams its own registered agents and MCP servers: opt-in, scope-guarded, synthetic payloads, no external egress.
Goal hijacking redirects an AI agent's objective while it runs normally. How ASI01 attacks work and the controls that contain a hijacked goal.
Rotation that has never been exercised is not rotation. Learn the overlap-window pattern, why agent fleets break rotation differently than services do, and how to make it routine rather than an incident.
Why point-in-time authorization fails for long-running AI agents, and how continuous authorization contains privilege abuse — OWASP ASI03.
Memory poisoning implants malicious content in an AI agent's persistent state. How ASI06 attacks persist across sessions and the defenses that work.
A registry is where MCP server choices get made, which makes it a supply-chain control point. Learn what a registry entry should carry, how ratings get gamed, and why internal catalogues beat public ones.
The AI agent supply chain — models, MCP servers, tools, prompts, frameworks — and the vetting, pinning, and attestation controls against ASI04.
Organizations have human offboarding checklists and nothing equivalent for agents. Retired agents keep credentials, connections, and scheduled triggers. Here is the deprovisioning sequence that actually closes them.
Agent-to-agent channels are an attack surface: spoofed peers, injected payloads, unbounded delegation. Securing A2A against OWASP ASI07.
Static allow-lists gate identity; dynamic trust scores gate scope. Learn how each works, where each falls short, and why mature programs combine both.
How one agent's error propagates through orchestrated systems — amplification, feedback loops, resource storms — and the ASI08 containment controls.
Rogue agents run with valid credentials outside governance. The kill-switch architecture — containment tiers, revocation, drills — for OWASP ASI10.
Developers add MCP servers to their agents in minutes, without review. Learn how to find the ones already running in your environment and how to make the governed path easier than the ungoverned one.
Guardrails, evals, and monitoring each close a different AI safety gap at a different lifecycle stage — learn how to use all three correctly.
Machine identity, workload identity, and agent identity are not synonyms. What each term means, where they overlap, and what agents add.
Non-human identity credentials from issuance to revocation: provisioning, scoping, rotation, monitoring, and decommissioning without sprawl.
A marketplace that lets publishers approve their own listings has no review at all. Learn the listing lifecycle, why self-certification must be structurally impossible, and what review should actually check.
The Nov 2025 MCP authorization flow, step by step: protected-resource metadata, client registration, PKCE, and RFC 8707 resource-bound tokens.
Third-party MCP servers are supply-chain risk: registry provenance, tool-description audits, version pinning, and an approval workflow that scales.
Sending every request to your best model is the most expensive way to run an agent. Learn the routing strategies that hold quality, the ones that quietly cost more, and how routing becomes a governance control.
MCP connects agents to tools; A2A connects agents to each other. Understand how both protocols divide labor and what security controls each one demands.
