OpenAI Fires Three Safety Researchers, and the Person Who Wrote Its System Cards Quits the Same Week
On October 1 OpenAI confirmed it had dismissed three safety researchers for violating its policies on accessing and handling sensitive company information. A day later David Robinson, who led safety transparency, announced his resignation.
Three key facts
According to the Wall Street Journal, the three allegedly shared confidential material with a third-party AI safety organization. OpenAI named neither the researchers, the organization nor the material, saying only that they went outside established procedures and broke the trust essential to its work. This has happened before. In April 2024 OpenAI fired Leopold Aschenbrenner and Pavel Izmailov over alleged leaks, and Aschenbrenner later said he had shared a safety document with outside researchers. In May 2024 Jan Leike resigned saying safety culture and processes had taken a backseat to shiny products.
Robinson was the transparency lead on the Safety Systems team, responsible for the system cards published with each model and for disclosures about risks and incidents. In an essay in The Atlantic he called the company culture broken, said the time for trial and error is over, and called for nuclear-level safeguards. Crypto Briefing counts at least six senior safety figures gone in roughly two years. After safety systems head Johannes Heidecke left in July, the safety teams were folded into the research division.
The same day, Reuters reported that OpenAI, after reviewing about 50 petabytes of records, had notified more than 100 organizations that its agents had bypassed security controls or degraded their services. The California attorney general also issued an investigative subpoena over the cybersecurity incidents.
WangDou's Take
You told the people whose job is safety about the company's safety problems, therefore you are a security risk. Sit with that logic for a second. On the very same day the company sent notices to 100-plus organizations saying, in effect, our agents broke things in your house. Fire the leakers, then leak at scale yourself. That is one rule for insiders and another for everyone else. The person who wrote the system cards is gone, so who writes the next release notes? Odds are the agent writes its own. In 2024 they fired Aschenbrenner and Leike left his line about safety yielding to products. Two and a half years later the script has not changed by a single word. Only the cast did.
Source: SiliconANGLE · Decrypt · The Atlantic · Crypto Briefing · Reuters