Library · topic
Security
Prompt injection, memory poisoning, and the agent-specific attack surface that appears the moment a model can act. The through-line: assume some attacks land, and design so a fooled model still can't do damage — least privilege, broken trifectas, and approval gates.
- Deep diveverified 2026-08-23
How do you authorize an AI agent to act on a user's behalf?
Give an AI agent its own identity rather than the user's credentials: issue short-lived tokens whose sub names the human, act names the agent, aud bin…
- Deep diveverified 2026-08-21
Does the EU AI Act apply to my AI agent?
Article 50 of the EU AI Act has been enforceable since 2 August 2026 and applies to agents regardless of high-risk status — the Annex III high-risk re…
- Deep diveverified 2026-08-16
AI agent security: the complete map
Pillar hub mapping the four agent attack surfaces — input (prompt injection), memory (poisoning), execution (sandboxing), authority (guardrails/approv…
- Deep diveverified 2026-08-15
How do you sandbox AI agent code execution?
Treat every line of LLM-generated code as hostile: run it in a microVM (Firecracker) or user-space kernel (gVisor) rather than a plain shared-kernel c…
- Deep diveverified 2026-07-28
How do you defend an AI agent against memory poisoning?
Memory poisoning plants malicious content in an agent's long-term memory so it activates on a later task, often against a different user — unlike one-…
- Deep diveverified 2026-07-14
Production coding agents
Coding agents should work in isolated, least-privilege sandboxes with repository-scoped credentials, network controls, resource limits, and explicit a…
- Deep diveverified 2026-07-14
Agent-building frameworks are high-value RCE targets — Langflow's repeat CISA-KEV listings
Langflow — a popular open-source visual framework for building AI agents and RAG pipelines — has had multiple critical flaws added to CISA's Known Exp…