Security

An AI reviewer's cleanup test read an awk array instead of testing membership, and kill -KILL took down 12 of my agents' sessions. The fix is a refusal.

Mini Shai-Hulud packages had valid provenance and were still malicious, and under EU AI Act Article 50, Code adherence is not conclusive evidence either.

Sleeper poisoning planted fabricated memories in up to 99.8% of attempts on GPT-5.5, so AI memory must prove it is current before it is obeyed.

45% of AI-generated code contains security flaws. 86% fails XSS. The illusion of correctness is the real risk — and it's hiding in plain sight.