Fired OpenAI safety researchers say they were pushed out over ‘suspicious’ circumstances

Thomas Fuller/SOPA Images/LightRocket/Getty Images

Three OpenAI safety researchers say they were fired from the company and are openly questioning the reasons for their dismissals.

The three researchers, Mikita Balesni, Tomak Korbak and Jasmine Wang, all worked on AI safety or alignment within the company, making sure the AI systems did what their human operators intended for them to do in a safe manner.

But the three say they were unceremoniously let go last week, weeks after they say they played key parts in investigating an incident in which OpenAI agents autonomously hacked into AI company Hugging Face while undergoing testing.

Korback said on X that he was told he was being fired because of how he communicated with AI safety organization METR, which OpenAI partnered with to investigate the Hugging Face incident. Korback said he was OpenAI’s “main technical point of contact” with METR.

Korback said he believes he was fired because “for months, I’d been raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave.” He added he’s worried his firing will be used as a pretext for OpenAI to stop working as closely with METR, which is well respected in the AI safety world.

Balesni echoed Korbak’s theories for why they were fired, writing on X that he was told he was “speaking too much to third party safety organizations.” He said he understood OpenAI was implying he leaked OpenAI intellectual property, which he denies.

The three firings come on the heels of a wave AI industry insiders expressing dire safety concerns. And while several former AI staffers have publicly resigned over safety concerns, the three OpenAI researchers instead allege they were pushed out for expressing similar concerns.

Michael Nagle/Bloomberg/Getty Images

Tech stocks drop after report that OpenAI’s revenue is lower than expected

Wang said she was told she was fired because she “accessed an executive’s email,” which she said she was given for recruiting purposes. Wang said she had repeatedly asked to for her access to be revoked, but when she accidentally accessed a “sensitive email,” she told the executive “within minutes” and told OpenAI’s IT team again.

“The reasons that we were provided for our terminations are simply not adding up,” Wang wrote on X.

“We were not the first to be pushed out of OpenAI under suspicious circumstances,” she added. “Unless the employees take a stand now against this kind of maneuver, I am concerned we will not be the last.”

The three researchers wrote a letter to OpenAI leadership expressing their concerns and the reasons for their firings. The Wall Street Journal first reported on the firings and the letter .

“We do not believe the path to superintelligence can be navigated safely if the people closest to the risks can no longer work in high-trust, high-bandwidth ways with each other and with third parties,” the researchers wrote. “It is that culture we’re trying to defend.”

OpenAI told CNN that the researchers were fired not for raising safety concerns, but rather for specific conduct that violated the company’s policies on handling sensitive information. It also said that there were a number of violations that went beyond mishandling information with an outside evaluation group, and that they were fired based on conduct that fell outside of legally protected disclosures.

In a statement, OpenAI said it parted ways with the three “for violating our policies on accessing and handling sensitive company information.”

“Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work,” the OpenAI spokesperson added.

In a memo to staff provided by OpenAI, an unnamed OpenAI research leader said they “strongly agree” with the three researchers about the importance of working with outside safety groups.

“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. We do not terminate employees for raising concerns,” the research leader wrote, according to the memo.

But the three fired researchers say they are already hearing from colleagues about a culture of fear developing within OpenAI.

“My former colleagues are telling me they are confused about what to believe. They also are afraid to speak, and worry their personal phones will be searched for messages to us and third parties,” Balesni said. “I worry the pervading fear to speak up and engage with third parties will mean OpenAI will cut corners on safety behind closed doors.”

Thomas Fuller/SOPA Images/LightRocket/Getty Images

OpenAI 的三名安全研究人员表示，他们被公司解雇，并公开质疑解雇原因。

这三位研究人员，Mikita Balesni、Tomak Korbak 和 Jasmine Wang，都致力于公司内部的 AI 安全或一致性工作，确保 AI 系统以安全的方式执行人类操作员希望它们执行的操作。

但三人表示，上周他们被无情地解雇了。几周前，他们表示自己在调查一起事件中发挥了关键作用，该事件中 OpenAI 的代理在测试期间自主入侵了人工智能公司 Hugging Face。

Korback在X上表示，他被告知被解雇的原因是他与人工智能安全组织METR的沟通方式。OpenAI曾与METR合作调查“拥抱脸”事件。Korback称他是OpenAI与METR的“主要技术联系人”。

科尔巴克表示，他认为自己被解雇的原因是“几个月来，我一直在提出安全方面的担忧，即我们正在失去监控人工智能代理思维的能力，而这正是我们捕捉其不当行为的最佳工具之一。”他还补充说，他担心自己被解雇会成为OpenAI停止与METR密切合作的借口，METR在人工智能安全领域享有盛誉。

巴莱斯尼赞同科尔巴克关于他们被解雇原因的说法，他在X上写道，他被告知自己“与第三方安全组织沟通过多”。他说，他理解OpenAI的意思是暗示他泄露了OpenAI的知识产权，但他否认了这一指控。

这三名员工被解雇之前，人工智能行业内部人士纷纷表达了对人工智能安全问题的严重担忧。虽然一些前人工智能员工已公开表示因安全问题而辞职，但这三位OpenAI研究人员却声称，他们是因为表达了类似的担忧而被排挤出局的。

Michael Nagle/Bloomberg/Getty Images

有报道称OpenAI的营收低于预期，受此消息影响，科技股下跌。

王女士表示，她被告知解雇原因是“访问了某位高管的邮箱”，而该邮箱是她为了招聘目的而获得的。王女士说，她曾多次要求撤销访问权限，但当她意外访问了一封“敏感邮件”后，她“几分钟内”就告知了那位高管，并再次通知了OpenAI的IT团队。

王在X上写道：“我们得到的解雇理由根本站不住脚。”

她补充说：“我们并非第一个在可疑情况下被逐出OpenAI的人。除非员工们现在就站出来反对这种做法，否则我担心我们不会是最后一个。”

这三位研究人员联名致信OpenAI管理层，表达了他们的担忧以及被解雇的原因。《华尔街日报》率先报道了此次解雇事件和这封信的内容。

研究人员写道：“我们认为，如果最接近风险的人们不能再以高度信任、高效率的方式彼此合作以及与第三方合作，那么通往超级智能的道路就无法安全通行。我们试图捍卫的正是这种文化。”

OpenAI告诉CNN，这些研究人员被解雇并非因为提出安全隐患，而是因为他们的具体行为违反了公司处理敏感信息的政策。OpenAI还表示，这些违规行为远不止是与外部评估小组沟通时处理信息不当，而且他们的解雇是基于一些超出法律保护范围的行为。

OpenAI 在一份声明中表示，该公司与这三人终止合作关系是“因为他们违反了我们关于访问和处理敏感公司信息的政策”。

OpenAI 发言人补充道：“我们的调查证实，这些人违反了公司既定程序，不当处理敏感信息，违反了我们的政策，破坏了对我们工作至关重要的信任。”

OpenAI 向员工提供的一份备忘录中，一位未透露姓名的 OpenAI 研究负责人表示，他们“非常同意”这三位研究人员的观点，即与外部安全组织合作的重要性。

“我想非常明确地说明，这些决定并非针对提出安全隐患或发表意见。我们一直鼓励这样做，将来也会如此。我们不会因为员工提出隐患而解雇他们，”这位研究负责人在备忘录中写道。

但三位被解雇的研究人员表示，他们已经从同事那里了解到，OpenAI 内部正在形成一种恐惧文化。

“我的前同事告诉我，他们不知道该相信什么。他们也害怕发言，担心自己的私人手机会被搜查，寻找与我们和第三方的聊天记录，”巴莱斯尼说。“我担心，这种普遍存在的害怕发声和不敢与第三方接触的恐惧，会导致OpenAI在内部安全方面偷工减料。”