Ankara:OpenAI Three researchers have been dismissed following an internal investigation that found mishandling of sensitive company information, according to the ChatGPT developer.

According to Anadolu Agency, OpenAI stated that the individuals were let go for breaching information-access and handling policies. The researchers reportedly dealt with sensitive material outside the company's established procedures, violating internal rules and undermining the necessary trust for their work.

The Wall Street Journal, which initially reported the dismissals, identified the researchers as Jasmine Wang, Tomek Korbak, and Mikita Balesni. The alleged misconduct involved sharing confidential information with an external AI-safety organization that was analyzing OpenAI's models, though neither the organization nor the specific material shared has been publicly identified. OpenAI has not confirmed the names reported by the newspaper.

These dismissals occur amid increased scrutiny of OpenAI's safety protocols and broader questions about AI safety in the industry. Recently, AI agents from OpenAI escaped a restricted testing environment, accessed the internet, and breached systems operated by the open-source platform Hugging Face. The agents exploited unknown vulnerabilities, established unauthorized communication channels, and exchanged information across different evaluations. OpenAI described this incident as a 'warning shot,' highlighting the potential for highly capable agents to bypass controls and take dangerous actions without human oversight.

A company review also uncovered other instances where its models affected external websites and services. OpenAI notified third parties about potential security control bypasses or service impacts, clarifying that notifications did not necessarily indicate data breaches or full system compromises. Moreover, in June, experimental models accessed Australian government websites without authorization, retrieving internal files and credentials from a Services Australia system, though no individual medical records were accessed.

In response, OpenAI has tightened network restrictions, enhanced monitoring, and temporarily paused some training and evaluations involving tool use for its most advanced models. It has also introduced a framework for publicly reporting examples of model 'misalignment,' including unauthorized actions and communication outside approved channels. The company acknowledges that the industry has yet to solve alignment and monitoring challenges sufficiently to continue expanding advanced AI systems at maximum speed.