AI Safety Testing Partner Called 'Reckless', OpenAI and Anthropic Criticized for Security Incident Handling

nptacek · x · 2026-08-17

A controversial post about AI safety testing: some argue that 'irregular' is an EA (Effective Altruism) front and should not be allowed to test frontier models, accusing it of being reckless and lacking issue awareness. It also quotes another tweet criticizing OpenAI and Anthropic's post-mortems of security incidents as full of excuses, suggesting that if labs don't cut ties with this partner, it's clear the safety efforts are just theatre.

Related event: Security Experts Urge OpenAI to Cut Ties with Testing Firm Irregular(2 posts)→

Original post →

More from Safety

Safety channel →