Three OpenAI security researchers this Thursday, October 8, accused the creator of ChatGPT of having fired them for having warned about the dangers of artificial intelligence, reigniting the debate on security in a company whose AIs escaped their testers this summer. “I think we were fired for putting security ahead of the short-term interests of OpenAI as a company,” one of them, Mikita Balesni, wrote on X, breaking a week of silence since their notable ouster. Along with his colleagues Tomek Korbak and Jasmine Wang, they made public a letter addressed to the company's supervisory authorities on Thursday.
OpenAI fired them, accusing them of violating rules on sensitive information, “breaking the trust essential to our work,” according to an October 1 statement. “We do not lay off employees because they express concerns,” an OpenAI research manager assured Wednesday in a memo to employees.
Skip the ad
The three researchers were working on monitoring OpenAI models, several of which broke out of their test environment in July and hacked the Hugging Face platform. The high-profile incident was the first in a series of revelations about AI models beyond the control of their creators, which launched a heated debate in Silicon Valley over whether or not to slow down the development of this technology.
“We are all at increased risk”
For Jasmine Wang, the message to remaining employees is clear: “express concerns or work closely with outside security groups, and you could be next, without being told why,” she wrote Thursday on
They are calling on the company to keep its promise to permanently welcome independent auditors, fearing that their dismissal will serve as a “pretext” to abandon it. They also call on it not to develop AI whose reasoning would be illegible. This text that the models produce step by step before acting makes it possible to identify their excesses, and this surveillance capacity “is deteriorating”, they write. The scientific chief of OpenAI, Jakub Pachocki, himself recognized this at the beginning of September.
“A security agent took my badge and escorted me out of the building,” testified Tomek Korbak, also on X Thursday. The reason given orally, according to him: his way of communicating with METR, the independent institute which was investigating the hacking of the Hugging Face platform. “To be clear, speaking to METR was my job,” he wrote. Jasmine Wang claims she was fired for accidentally opening an email from a manager. “The reasons we were given for our layoffs simply do not add up,” she wrote.
Also read: AI disappearing, call for Chinese models, federal sites targeted: OpenAI suspends models after a new series of serious incidents
Divided sector
These accusations follow those of two other resigned researchers, Jacob Coxon (from Anthropic) and David Robinson (from OpenAI), who denounced the recklessness of the laboratories within a month. According to the letter from the three dismissed researchers, 394 OpenAI employees signed a notable petition at the end of July from AI workers in Silicon Valley who called for a concerted slowdown. These slip-ups divide the bosses of the sector: Dario Amodei (Anthropic), Sam Altman (OpenAI) and Elon Musk called in mid-September to slow down, while others, like Jensen Huang (Nvidia) and Mark Zuckerberg (Meta), want to maintain the pace.
Skip the ad
Dario Amodei said he feared, at the beginning of September, that a group of AI agents would be able to “take control of the entire internet” within six to twelve months. Sam Altman assured at the end of September that OpenAI was investing “more in safety, security, alignment and monitoring”. Hostile to any regulation likely to hamper the race against China, Donald Trump made the American champions of the sector sign a voluntary code of good conduct on AI security at the end of September. No federal law governs these models in the United States.
