Anthropic published a third-party result of zero successful prompt injections across 720 attempts against Claude Code auto mode. Six weeks later Johann Rehberger achieved remote code execution in 3 to 4 runs out of 5. Neither number is wrong, and the gap between them is a controls lesson every finance team already knows.
Check Point spent a year auditing LangChain, LangGraph, CrewAI, AutoGen, Microsoft Agent Framework, and Google ADK. They found the same 20-year-old bug classes in all of them. The lesson: assume prompt injection succeeds, and harden what happens next.
ChatGPT Atlas stops working August 9 after 9.5 months. Not a failure story — a story about what an agent becomes, and how little notice you get.
Cisco scanned the #1-ranked skill in OpenClaw's community repo and found nine security issues. Installing a skill is installing instructions.
The actual operational security rules FRED lives by after four months in production, written as a practical guide you can adapt for your own agent.
Simon Willison's lethal trifecta — private data, untrusted content, external communication — explains why AI agents get compromised, and what to do.
A fake bug report hijacked coding agents at a Fortune 100 company. OWASP says prompt injection may be permanent. The security map just got redrawn.
The most important question to ask before installing any AI tool — and the narrow-perimeter security architecture behind FRED.
Most AI agent setups add security later. Matt built FRED the opposite way — security first, capabilities second. The philosophy, in his own words.