AI Safety Researcher Calls OpenAI and Anthropic's Safety Evaluations "Fake"
peterwildeford · x · 2026-11-20
AI safety researcher Peter Wildeford called OpenAI and Anthropic's current safety evaluations "fake" on stage at MTS: labs get to tell evaluators what to inspect, redact the findings, and simply ignore recommendations.
He argued for independent evaluators modeled on banking and financial services regulators who actually scrutinize companies. He cited a framework released the same day by Representatives Jacobs, Beyer and Maloy proposing a government body that could set binding safety standards, with examiners potentially embedded inside AI companies.
More from Safety
- South Korea education minister resigns over AI cheating scandal engulfing suneung exams — nordicinst · 2026-11-22
- AI Detector Flags Researcher's 2016 Paper as 79% AI-Generated — AntonioCasilli · 2026-11-22
- Lean community certifies Riemann proof; honeypot fake keys expose breaches at three AI providers — LeviTurk · 2026-11-22
- Intruder swapped Prime Intellect's hosted DeepSeek for an abliterated copy for 53 hours — mawcopolow · 2026-11-22
- Dragos finds three hacked AI hosts whose operators had no idea they were breached — mawcopolow · 2026-11-22
- Who do you sue when an AI agent errs? The "sovereign agents" liability debate — QuixiAI · 2026-11-22