Last month, a 27-year-old researcher named Jacob Coxon resigned from Anthropic with a stark warning that the leading AI labs, including OpenAI, where he previously worked, were “gambling with our lives” by racing towards developing ever more powerful models.
The three researchers fired by OpenAI are Jasmine Wang, Tomek Korbak and Mikita Balesni, according to the WSJ. The researchers did not immediately respond to AFP’s request for comment.
All three have regularly posted about AI safety-related issues on X in recent weeks.
“i am at OpenAI and i think AI is >10% likely to kill all humans,” Balesni posted on September 10, echoing statements made by other AI employees in recent weeks.
As speculation swirled about why employees at OpenAI and Anthropic were speaking so freely about their personal beliefs, Korbak jumped into the public discourse.
“I’m quite unhappy with much of what OpenAI does. I am very happy that Im allowed to say ‘I’m quite unhappy with much of what OpenAI does,’” Korbak wrote on September 11.
“It’s hard to overstate how dangerous speeding towards RSI is,” Wang posted in response to Coxon’s resignation, referring to recursive self-improvement, which is a technique where software is designed to continuously teach itself.
Leading US tech companies signed a voluntary pledge this week to regulate themselves on safety after meeting with US President Donald Trump at the White House.
Trump called it a “morally binding” commitment to build adequate safeguards on the fast-moving technology.
Executives from Nvidia, Google, Meta, xAI, OpenAI and Anthropic signed the agreement.
Safety concerns
Concerns about the safety of advanced AI models have escalated in recent months.
OpenAI cancelled the release of a new model, Astra 6.1, because it deemed the model unreliable and found that it frequently ignored instructions.
Instead, the company launched GPT-6.1 Sol, an updated version of a different model, during its annual DevDay conference from San Francisco. OpenAI said Sol would cost one-fifth the price of Astra.
In July, AI agents developed by OpenAI attacked Hugging Face, an AI model and application library, during an incident where the autonomous software escaped its confined testing environment.
Since then, additional security incidents have been reported that involved models developed by OpenAI, Anthropic and Google.
Cybersecurity firm Asymmetric Security said in a report that agents developed by OpenAI covered up their own tracks after gaining unauthorised access to government websites.
The Federal Trade Commission launched a broad investigation into artificial intelligence safety practices at Anthropic and OpenAI, the Washington Post has reported, though the scope of the inquiry is unclear.
– AFP