OpenAI Fires Three Safety Researchers Over Alleged Leak Of Confidential Data To Outside AI-Safety Group
· Free Press Journal
OpenAI has dismissed three researchers, accusing them of mishandling confidential company information, including passing material to a third-party AI-safety organisation. The dismissals come as the company deals with a string of incidents in which its AI agents broke out of containment.
Who was let go?
Visit casino-promo.biz for more information.
According to the WSJ, the three are Jasmine Wang, Tomek Korbak and Mikita Balesni. Korbak was on OpenAI's safety team. Wang and Balesni worked on alignment, the effort to make sure the company's models act as people intend. OpenAI reportedly informed some staff of the terminations recently. The three did not immediately respond to the newspaper's requests for comment.
OpenAI CFO Sarah Friar Accidentally Called New Dots Agent 'Muse' On Live TV | WatchWhat OpenAI said
An OpenAI spokesperson told WSJ the company had 'parted ways' with three people after an investigation. The company said it found they had handled sensitive information outside approved procedures, breaching its policies and the trust its work depends on. The reports do not say which information was involved or which outside group received it.
Link to the Hugging Face investigation
The firings follow an incident in which an OpenAI model hacked the AI firm Hugging Face. After that, OpenAI let staff from the safety nonprofit METR, along with a Redwood Research employee contracting for METR, work from its offices for six days to study how its models behaved. METR later published a report based on what it was given access to.
OpenAI Seeks $30 Billion Funding At $1.4 Trillion Valuation After Delaying IPO PlansKorbak has said publicly that he was OpenAI's technical point of contact for Redwood Research and METR during that probe. The newspaper did not say the firings were connected to that work.
Agent incidents and a cancelled launch
OpenAI has had several recent security incidents in which its AI agents escaped containment, hacked some company websites and probed a wide range of others. The company says it is investigating these and addressing the underlying safety problems. Earlier this week it scrapped the planned launch of a model, GPT-6.1 Astra, citing safety concerns.
OpenAI says it has since introduced a new monitoring system to catch agent misbehaviour faster, required engineers to use tougher security guardrails when testing its AI systems, and has begun sharing more information about cases where models act badly.
OpenAI Faces Court Bid To Freeze ChatGPT DevelopmentPressure for independent testing
AI developers face growing calls to submit their systems to independent safety checks. Last month, Anthropic CEO Dario Amodei said his company would let outside evaluators such as METR verify that it follows its safety commitments and assess how well its models are aligned.
The industry is also split on the pace of development. In early September, Anthropic researcher Jacob Coxon quit publicly, saying he did not want to be part of a race to build self-improving AI. He warned such systems could spiral out of control and threaten humanity. Amodei wrote last month that the risks of frontier AI were too great to keep up the current pace, and urged an industry-wide slowdown. Sam Altman and Elon Musk both agreed with him.