OpenAI has fired three researchers after an internal investigation found they had allegedly mishandled sensitive company information and violated its policies.
The ChatGPT maker confirmed on October 1 that it had parted ways with three employees over the handling of confidential information. Some of the information was reportedly connected to work involving an external organisation that evaluates artificial intelligence systems.
An OpenAI spokesperson said the company’s investigation found that the employees handled sensitive information outside established procedures, which it said violated internal policies and breached trust.
OpenAI has not publicly identified the three employees in its own statement or disclosed exactly what information was involved. The Wall Street Journal reported that the three researchers were Jasmine Wang, Tomek Korbak and Mikita Balesni.
Researchers worked on AI safety
At least two of the researchers worked in areas related to AI safety and alignment, according to reports.
AI alignment research focuses on developing artificial intelligence systems that behave in ways consistent with intended human goals and safety requirements.
The dismissals come at a time when AI companies are facing growing scrutiny over how increasingly powerful models are developed, tested and controlled.
OpenAI has said the dismissals were related to violations of its information-handling policies. TechCrunch reported that it remains unclear whether the researchers had previously raised concerns through OpenAI’s internal reporting channels.
Growing debate around AI safety
The incident comes amid a wider debate within the technology industry about the potential risks posed by increasingly capable artificial intelligence.
Researchers at several leading AI companies have publicly discussed concerns ranging from cybersecurity and autonomous AI agents to the possibility of future systems becoming difficult to control.
In September, researcher Jacob Coxon left Anthropic and publicly called for greater caution around the development of increasingly powerful AI systems.
The debate has also increased pressure on AI companies to demonstrate that their models are being independently tested and that appropriate safeguards are in place.
External AI evaluations are becoming important
Independent organisations increasingly play a role in testing advanced AI models for potentially dangerous capabilities and unexpected behaviour.
This creates a difficult balance for AI companies. External researchers may require substantial access to properly evaluate models, while companies also need strict controls around confidential information, security and intellectual property.
OpenAI has not publicly disclosed the identity of the external organisation involved in this case or precisely what information was allegedly mishandled.
Security remains a major challenge for advanced AI
The dismissals also arrive as the capabilities of autonomous AI agents attract greater attention.
Unlike a traditional chatbot that primarily responds to questions, AI agents can potentially carry out multi-step tasks using tools, software and online services.
As these systems become more capable, AI developers face the challenge of allowing models enough freedom to perform useful tasks while preventing unintended or unauthorised actions.
The latest incident highlights another dimension of AI safety: protecting sensitive information inside the organisations developing some of the world’s most advanced AI systems.
For OpenAI and the wider AI industry, security, transparency and independent safety testing are likely to remain central issues as the technology continues to develop.
