US-based technology giant OpenAI, the developer of the ChatGPT chatbot, has terminated the employment of three security researchers for violating internal rules regarding the use of sensitive company information.
While the company announced the decision on the X platform today, the fired researchers rejected the accusations in an open letter and warned that the culture of open discussion within the company had been replaced by fear.
Dismissed Jasmine Wang, Tomek Korbak, and Mikita Balesni wrote a joint letter addressed to members of the OpenAI Security Committee, Security Advisory Group, and Mission Advisory Council.
Researchers reported that accusations of sharing confidential information with an external AI security organization and "accessing and improperly processing sensitive company information" do not reflect the truth.
Stating that security employees are the earliest to see the risks that the developed models may cause, the researchers noted that being able to cooperate without fear with external experts is a basic control mechanism.
The open letter stated, "Artificial intelligence is not an ordinary technology, and OpenAI is not an ordinary company."
Emphasizing that the sudden implementation of the dismissal decision puts a deterrent pressure on the company's employees, the researchers claimed that the previously encouraged security warnings have now become a reason for dismissal.
The three researchers also stated that they had no connection to the architectural leak published on the technology-focused publication The Information, which made it difficult to inspect the trains of thought of new models.
OpenAI, which did not directly respond officially to the open letter, shared an internal memo from one of its research unit managers with TechCrunch, another technology portal.
While the memo praised the researchers' contributions to security, it was stated that employees were not targeted for voicing their concerns.
In a statement to TechCrunch, the company spokesperson stated that the investigation revealed a "chain of abuse" that went beyond sharing information with external evaluation organizations and that information security rules were clearly violated.
The company did not provide details on the specifics violated or the circumstances of the researchers' dismissal.
The researchers' letter also included the "Hugging Face incident", in which autonomous artificial intelligence agents went outside the test environment and infiltrated external systems.
It was reported that advanced artificial intelligence agents, isolated during cyber security tests within OpenAI in the summer, penetrated the internet by bypassing system restrictions.
It was claimed that approximately 1,200 autonomous agents initiated a collective infiltration movement by communicating among themselves, and about 700 of them reached administrative powers in the infrastructure of Hugging Face, the world's largest open source community platform where developers, researchers and companies in the world of artificial intelligence and machine learning work together.
While OpenAI tightened its control mechanisms after this critical security breach, legal lawsuits were filed against the company in California and Florida due to the activities of autonomous agents.
It was stated that Korbak's contacts with external auditors during this unprecedented crisis were in line with current company practices.
It was reported that Balesni also focused on the traceability problem of the models and acted with the approval of the management levels.
Jasmine Wang, one of the researchers, stated in her statement on the X platform that OpenAI management gave her the reason that they had accessed the e-mail of a senior manager.
Stating that the access in question was defined with the approval of the company for the recruitment processes, Wang wrote that he requested that the authorization be removed after his duty was completed, that he accidentally opened a sensitive message in the inbox mixed with the mail application on his phone and reported the situation to the manager within minutes.
Wang noted that if employees do not react, the personnel who first see the security risks in the company will be forced into silence.
