A debate is playing out on social media, and inside the walls of OpenAI, about why the company fired three individuals last week on its safety team. Now, the former employees, as well as OpenAI, have issued additional statements to clarify their sides of the story.
Two of the fired researchers—Tomek Korbak and Mikita Balesni—worked on OpenAI’s investigation of the Hugging Face incident. Korbak was the technical point of contact for METR, the third-party research firm that OpenAI asked to conduct an independent audit of the incident. The third dismissed employee, Jasmine Wang, was a program manager on OpenAI’s safety team.
The Wall Street Journal first reported the dismissals on Oct. 1. At that time, OpenAI said the “individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
Fired employees publish a letter responding to the allegations
Since then, there appears to have been continued conversation and controversy among the ranks at OpenAI, prompting the three former employees to co-author a four-page letter, published on Oct. 8.
The letter responds to “various versions of events circulating” among OpenAI staffers, and refers to ongoing “internal and external communications” about the firing that the former employees say is making current employees “afraid to speak” up.
“Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prized in the past,” the letter says. “It is that culture we are writing to defend.”
The two biggest themes in the letter are about OpenAI’s ability to monitor its advanced models, and about its engagement with third party safety companies, such as METR. The former employees deny they leaked concerns about the ability to monitor OpenAI’s latest Astra model to The Information, which published a Sept. 1 article about it. Second, they deny any foul play in their communication with METR. The staffers urge OpenAI to preserve the ability to closely monitor its advanced models, to adhere to its commitment to embed third-party safety auditors within the organization, and to foster an open and transparent culture.
“We are concerned that our firings may be used to justify ending OpenAI’s work with METR, or otherwise providing external auditors much more limited access and scope,” the letter says.
Varying accounts of reasons for the firing, but nothing concrete
OpenAI has not disclosed the specific reasons for the firing, but told Fortune it found a pattern of misconduct from the three employees, including a number of violations, which clearly violated its policies for handling information. It did not link the firings to its communication with METR, or the Hugging Face incident.
Each former employee gave their own brief explanation this week in social media posts that link to the joint letter.
Korbak said OpenAI told him verbally, not in writing, that he “was fired because of the way I communicated with METR. “Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took my badge and walked me out of the building,” Korbak said. “Talking to METR was my job.”
Wang said the reason OpenAI gave her was that she accessed an unnamed executive’s email. She said the company gave her that access for recruiting, and when she no longer needed it, she asked IT to remove it, but they did not complete her request.
“We were not the first to be pushed out of OpenAI under suspicious circumstances,” Wang said. “Unless the employees take a stand now against this kind of maneuver, I am concerned we will not be the last. The message to everyone still at OpenAI is clear: raise concerns or work closely with outside safety groups, and you could be next, without being told why.”
Balesni said they “were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.”
Does OpenAI have an unhealthy safety culture?
In response to the letter, OpenAI sent an internal memo to employees, as well as put out a public statement. “Our internal investigation uncovered a significant breach of trust beyond what’s outlined in the letter they published and we stand by the decision to not continue their employment,” the company said.
But lack of clarity about what exactly the employees did, or may have shared with METR, has fueled speculation about what happened. (METR did not respond to our request for comment.)
“I can’t really see a world where OpenAI’s actions are justified here,” said Neel Nanda, a Google DeepMind employee who was formerly at Anthropic. “Firing people over a good faith attempt to use their best judgement in a novel and uncertain situation is a sign of a highly unhealthy culture.”
Former OpenAI safety transparency lead David Robinson also spoke out this month about the company’s “broken” safety culture in an op-ed in The Atlantic, saying the company is not being “nearly safe enough” and relying on “trial and error.” He highlights the Hugging Face incident as evidence of this, and that the safeguards the company put in place in response still didn’t prevent more rogue incidents.
OpenAI shared a portion of its internal memo to employees, sent by its research leaders with Fortune—presumably penned by Mark Chen, Chief Research Officer, though the company did not specify. The memo affirmed the company’s commitment to working with third party safety assessors, and said “monitorability is of the utmost importance to us.”
“I cherish our open culture of debate, especially when it comes to safety, and consider this openness critically important to making the right decisions,” the memo said. ” I want to be very clear that these decisions were not about raising safety concerns or speaking out. We have always encouraged that and always will.”
OpenAI said it is still committed to working with third party safety assessors, and is finalizing contracts now that it will announce “in the coming weeks.”
This story was originally featured on Fortune.com
