Relying on OpenAI's 'Goodwill' for Safety Investigations Criticized
sjgadler · x · 2026-08-27
DKokotajlo comments on the dynamic between AI safety evaluators (like METR and Redwood) and OpenAI. He notes that evaluators may try to keep OpenAI happy to ensure continued access for future investigations. He argues this situation is terrible for humanity and that OpenAI should be legally required to open up to independent teams, rather than investigations depending on the company's goodwill. He clarifies his respect for METR and Redwood despite the systemic issue.
Related event: Experts Slam OpenAI Incident Probe as Too Narrow and Not Truly Independent(35 posts)→
More from Safety
- Beware of Phishing Attacks Impersonating Bloomberg on X — samuelcolvin · 2026-08-28
- Speculation: Will AI Swarm Write 'Straussian' Text with Dual Meanings? — soumitrashukla9 · 2026-08-28
- If models get a credit card to buy compute, hacking money is just social engineering — BLUECOW009 · 2026-08-28
- moyix to host talk on handling auth in autonomous web pentesting — moyix · 2026-08-28
- Shared Agent Skill Libraries Propagate Malware, 41.8% Self-Poisoning Rate Found — omarsar0 · 2026-08-27
- Matthew Green questions OpenAI security awareness — matthew_d_green · 2026-08-27