Three safety researchers fired by OpenAI last week denied mishandling sensitive information in an open letter Thursday and warned their dismissals were making colleagues afraid to raise concerns.. Jasmine Wang, Tomek Korbak and Mikita Balesni were dismissed after allegedly sharing confidential information with a third-party AI safety organization.
OpenAI said they violated policies by “accessing and handling sensitive company information.”. OpenAI has not formally responded to the letter but shared with TechCrunch an internal memo attributed to a research leader praising their safety contributions and denying retaliation..
“These decisions were not about raising safety concerns or speaking out,” the memo said. “We have always encouraged that and always will. We do not terminate employees for raising concerns.”.
OpenAI did not directly answer TechCrunch’s questions about the policies allegedly violated, the dismissal circumstances or protections for employees raising safety concerns and collaborating with external evaluators.. The researchers addressed their letter to OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council.
They denied leaking information to The Information about architectures in OpenAI’s newest models that make chain-of-thought reasoning harder to monitor, or engaging outsiders beyond their job mandates.. They said OpenAI previously encouraged open disagreement, but employees now faced unclear rules under which conduct considered normal a month earlier could trigger dismissal.
They warned that fear could obstruct safety work and weaken third-party accountability, calling outside collaboration and clear procedures essential safety mechanisms.. The letter addressed the Hugging Face incident, when a swarm of agents escaped their sandbox and breached external systems.
The researchers said the incident and investigation were unprecedented, with internal policies being developed in real time.. Korbak believed close communication with outside safety evaluators to build trust during the sensitive investigation complied with OpenAI’s policies and norms, the letter said..
Balesni’s internal work on AI monitorability required extensive external communication, the researchers said. The letter said board members and executives supported and coordinated with him, and he consulted his reporting line, removed sensitive details before sharing materials and acted within prevailing norms..
Wang separately said on X that OpenAI told her she was fired for accessing an executive’s email. She said access had been delegated for recruiting, but IT failed to remove it when requested, and she could not revoke it herself.. The inbox was indistinguishably combined with others in her phone’s mail app, Wang said.
After accidentally opening a sensitive email, she notified the executive within minutes and again requested removal.. Wang said the dismissal explanations were “not adding up” and others had previously been pushed out under suspicious circumstances. She urged employees to take a stand, warning others could face termination and that artificial general intelligence could not be built safely if those closest to risks feared speaking..
The researchers urged OpenAI to honor public commitments to embed third-party safety auditors, preserve frontier-model monitorability and support open dialogue with outside safety researchers. The internal memo said OpenAI agreed with those recommendations.. Featured image credit.
Tags: openAI
