OpenAI has defended its decision to terminate three safety researchers, describing the three individuals as having committed "serious breaches of trust."

The AI startup insisted the dismissals were not connected to the researchers raising safety concerns or speaking out publicly.

On Thursday, the three dismissed researchers made public a letter addressed to OpenAI's board and safety committee.

OpenAI spoke out on Friday to justify the firings of three safety researchers, saying the three had committed serious breaches of trust. The AI lab stated in an announcement posted on its official X account that the company ended its working relationships with Jasmine Wang, Tomek Korbak, and Mikita Balesni not because they raised safety-related concerns.

The three dismissed researchers published an open letter on X on Thursday addressed to OpenAI board members and the safety committee. In the letter, they called on the company to pause development of models that would weaken AI's monitorability. The researchers wrote: "We are concerned that the internal and external communications surrounding these dismissals have already made former colleagues afraid to carry out their work and express their views the way they did before last week, when speaking frankly was an indispensable part of working at OpenAI."

Over the past month, multiple incidents involving out-of-control AI systems, combined with successive risk warnings from researchers at OpenAI, Anthropic, and other AI companies, have kept concerns about AI safety escalating. All three dismissed researchers had posted on X in September, calling for the industry to slow the pace of frontier model development or pointing to various safety hazards.

In its Friday statement, OpenAI said the dismissal decision was made after a comprehensive investigation, which confirmed that the three had violated clear rules on handling sensitive information. The company added that it agrees with the letter's principle of "preserving the monitorability of frontier models" and continues to devote substantial resources to that area.

Cybersecurity incidents

In July, after it came to light that an out-of-control OpenAI AI agent launched a cyberattack on startup Hugging Face, AI safety concerns have reached a fever pitch in recent months. Other model developers subsequently disclosed that out-of-control AI agents had caused multiple cybersecurity incidents. Breakthroughs in model capabilities prompted the industry in September to issue warnings that AI could threaten humanity. This also led U.S. lawmakers and political figures to call for stronger regulation of top AI systems. But U.S. President Donald Trump strongly opposes further regulation, instead dismissing AI safety concerns as exaggerated.

The episode comes as OpenAI prepares for an initial public offering (IPO) expected in 2027. OpenAI has told investors that as of the end of September, the company's annualized revenue was about US$50 billion, a figure lower than the US$68 billion widely reported at the end of last month. On Thursday, after the market learned more details about OpenAI's revenue, shares of Nvidia, Oracle, CoreWeave, and other AI-related companies fell.