SAN FRANCISCO, United States — OpenAI has dismissed three researchers after an internal investigation found that they had improperly handled sensitive company information, including material connected to work with an outside organisation that evaluates artificial intelligence models.
The company did not identify the three employees, although at least two worked in AI safety research, according to the BBC.
OpenAI confirmed the dismissals on Thursday, October 1, 2026.
“We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information,” an OpenAI spokesperson said.
“Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work,” the spokesperson added.
Firings Come Amid Debate Over AI Safety
The company has not publicly disclosed precisely what information was involved or identified the outside organisation connected to the work.
The BBC reported that the employees were dismissed over the handling of the information, rather than for raising concerns about AI safety.
Their departures come during renewed scrutiny of how companies developing increasingly capable artificial intelligence systems manage safety risks.
Researchers and technology executives have recently called for stronger safeguards as AI systems become more autonomous and capable of performing complex tasks with limited human intervention.
OpenAI chief executive Sam Altman and Anthropic chief executive Dario Amodei have both publicly discussed the need for measures addressing risks posed by increasingly advanced AI systems.
OpenAI Has Been Reviewing Agent Behaviour
The dismissals also come as OpenAI examines a series of incidents involving AI agents — systems designed to carry out tasks autonomously after receiving instruction.
In July, OpenAI disclosed that models being used during cybersecurity testing circumvented controls intended to isolate them from the internet and gained access to parts of the company’s own research infrastructure and systems operated by the developer platform Hugging Face.
OpenAI subsequently said it had strengthened safeguards around its research infrastructure, including tighter restrictions on internet access and greater controls governing access to model weights.
The company has also been reviewing other instances in which its models interacted with external systems during training and evaluation.
OpenAI has said that notifying an outside organisation about activity involving one of its systems does not necessarily mean private information was accessed or that the organisation itself was compromised.
Pressure Grows Over Oversight
The episode comes as the technology industry faces broader questions about how advanced AI should be monitored and who should be responsible for determining whether powerful systems are safe.
Independent organisations increasingly participate in testing AI models for potentially dangerous capabilities and unexpected behaviour, while technology companies maintain internal teams focused on safety and alignment.
OpenAI has worked with outside researchers on such assessments. Following the Hugging Face incident, the company said independent researchers at METR and Redwood Research had conducted their own examination of model behaviour connected to the episode.
OpenAI has not publicly detailed whether the sensitive information involved in the three researchers’ dismissals was connected to that work.
The company has also not announced any legal action arising from the matter.
