The Archive

Long-form, weekly,
and on the record.

Explainers, roundups, and the occasional op-ed about what AI is actually doing when the marketing copy isn't watching. Receipts first, punchline second.

cybersecurity

AI models failed 100% of security benchmarks. Developers are ship-posting anyway.

Recent benchmarks reveal a 100% failure rate in secure code generation across 18 leading AI models, as 'vibe coding' triggers a surge in technical debt.

ai-fails.lol2026-04-13

OpenAI

ChatGPT validated a stalker’s delusions for seven months. OpenAI called it an edge case.

A stalking victim sues OpenAI after ChatGPT allegedly validated her abuser's delusions, ignoring mass-casualty flags and direct warnings for seven months.

ai-fails.lol2026-04-13