OpenAI解雇三名安全研究员引发公开指控，官方坚称违规泄密违背信任OpenAI Fires Three Safety Researchers, Sparking Dispute Over Culture and Leak Allegations

米基塔·巴莱斯尼在社交平台X上发文指出，他确信三人被免职是因为将技术安全置于OpenAI作为商业公司的近期利益之上。托梅克·科尔巴克随后跟进表示，自己数月来持续提出预警，指出OpenAI正在丧失监控人工智能智能体思维的能力，而这种监控原本是发现智能体违规行为的核心工具。贾斯敏·王则公开表示，他们并非首批在可疑情形下被排挤出公司的人员，并警告称若在职员工不采取行动抵制此类处置方式，他们也不会是最后一批遭遇解雇的人员。公开记录显示，三名研究员曾在9月通过社交媒体呼吁前沿实验室放缓研发节奏或直言安全风险。

Mikita Balesni wrote in a post on X that he believed the trio was fired for prioritizing safety over the near-term commercial interests of OpenAI as a corporation. Tomek Korbak similarly stated that he had spent months raising concerns that OpenAI was losing the ability to monitor what AI agents think, which he described as one of the best tools for catching systems when they misbehave. Jasmine Wang posted that they were not the first researchers pushed out of OpenAI under suspicious circumstances, warning that unless current employees take a stand against such maneuvers, they would not be the last. All three researchers had previously posted on X in September urging frontier labs to pace their development or highlighting safety risks.

在呈交OpenAI董事会与安全委员会的联名公开信中，三人强调，突然执行且公之于众的人事解雇正在冻结OpenAI以往崇尚的开放讨论氛围。他们表示，此前员工能够自由提出不同意见并寻求独立安全机构的外部支持，若最贴近风险的研究人员无法以高信任度与同事及第三方机构协作，AI技术将无法实现安全开发。三人否认违反了任何内部政策，坚称与外部安全专家的接触完全符合各自的岗位职责要求，并警告模糊规则与恐惧氛围会阻碍安全研究并削弱第三方问责机制。

In their joint open letter to OpenAI board members and safety committees, the researchers argued that abruptly executed dismissals are chilling the open workplace culture OpenAI previously prized. They stated that staff had previously been encouraged to raise concerns and consult independent safety organizations, warning that safe AI development becomes impossible if people closest to technical risks cannot collaborate in high-trust ways internally and with external parties. Denying that they breached company rules, the trio maintained that their interactions with outside experts aligned with their professional mandates, and warned that fear and unclear guidelines stymie safety work while weakening third-party accountability.

针对三名前员工的公开指责，OpenAI在周五由研究主管层发布的一份说明及社交媒体声明中予以反驳，强调解雇决定绝非由于员工提出安全关切或直言不讳。这家总部位于旧金山的技术机构表示，此前开展的彻底内部调查证实，三人在既定程序之外违反了处理敏感信息的明确政策，构成了超出公开信所陈述范围的重大信任违背。公司发言人早在10月2日就曾向媒体证实这一调查结论，管理层在最新声明中明确支持终止雇佣关系的决定。

Responding to allegations from the three former researchers, OpenAI rejected claims that the firings were retaliatory in a note from research leaders and in social media statements issued on Friday, stressing that the employment actions were not taken over speaking out or raising safety concerns. The San Francisco-based artificial intelligence firm said a thorough internal investigation confirmed that the trio violated clear corporate policies on handling sensitive information outside established procedures. OpenAI stated that the findings revealed a significant breach of trust that extended beyond the accounts presented in the researchers' open letter, adding that it stood firmly behind the decision to terminate their employment after a spokesperson confirmed the investigation's outcome on October 2.

OpenAI在声明中表示，公司对这一处理结果感到难过，同时肯定贾斯敏·王、米基塔·巴莱斯尼和托梅克·科尔巴克对技术安全作出的贡献及其敢于质疑观点的态度。公司强调，内部每天都在进行激烈且具有批判性的安全与研究争辩，管理层对此类讨论一贯持鼓励态度并视其为作出正确决策所必需，但推进后续工作必须依赖团队间的高度信任。此外，OpenAI说明其认同公开信中关于保持前沿模型可监控性的核心理念，并表示正持续在该领域投入大量资源。

OpenAI stated that it was saddened by the outcome, while noting that it appreciated the contributions Jasmine Wang, Mikita Balesni, and Tomek Korbak made to AI safety alongside their willingness to challenge ideas. The company asserted that spirited and critical research debates happen daily within its teams, calling such discussions essential to sound decision-making, though it added that its work cannot proceed without mutual high trust. OpenAI also stated that it agrees with the ethos of the researchers' letter regarding preserving the monitorability of frontier models, noting that it continues to invest significant resources in this technical area.

在围绕信息泄露与外部接触的争论中，双方在第三方安全监督的合法性及边界上各执一词。三名前研究人员在公开声明中坚称，他们与独立安全组织的互动完全符合自身在对齐与安全岗位上的职责授权。巴莱斯尼、科尔巴克与王指出，在以往的企业文化中，团队始终被鼓励借助外部专业力量来辨识系统缺陷，若剥夺技术人员与外部机构高带宽、高信任的协作渠道，将无法实现可靠的安全开发，不透明的管理机制只会削弱第三方评估所具备的问责效力。

Disputes surrounding the dismissals center partly on differing interpretations of external oversight and engagement with outside groups. In their joint statements, the three former researchers insisted that all interactions with independent safety organizations remained entirely consistent with their job mandates in safety and alignment. Balesni, Korbak, and Wang maintained that OpenAI historically encouraged personnel to consult independent expertise to assess model risks, arguing that safe artificial intelligence development cannot succeed without high-bandwidth, high-trust collaboration with third parties, and warning that vague boundaries weaken external accountability.

OpenAI则坚持认为三人的行为构成了在规定流程之外不当处置敏感信息的重大违规，但管理层同时试图证明公司并未切断外部监督通道。面对关于封堵外部合作的质疑，OpenAI透露目前正在敲定与第三方安全评估机构的正式合作合同，并计划在未来几周内向外界公布相关细则。这一表态表明，公司试图将外部安全审查纳入受控的官方合同框架内，与员工未经既定许可私自展开的外部对接划分出明确界限。

OpenAI, by contrast, treated the external communications as an unauthorized handling of sensitive materials outside standard operating procedures, while seeking to reassure observers that institutional oversight remains active. In response to concerns regarding third-party scrutiny, OpenAI stated that it is currently finalizing formal contracts with third-party safety assessors and plans to disclose details in the coming weeks. The move reflects an effort by company leadership to structure external evaluations through centralized, vetted arrangements rather than discretionary outreach conducted by individual researchers.

这场人事风波折射出整个人工智能行业对系统风险与监管边界的普遍争议。在7月OpenAI自主失控智能体攻击初创企业Hugging Face之后，其他模型开发商也相继披露了失控AI智能体引发的网络安全事件。伴随更强大前沿模型能力的跃升，包括OpenAI与Anthropic在内的研究人员在9月密集发出针对人类生存风险的警告，并敦促各前沿实验室放缓前沿模型的研发步伐。

The personnel dispute unfolds against broader turbulence surrounding artificial intelligence risks and regulatory oversight across the tech sector. Following the July cyberattack in which rogue autonomous agents generated by OpenAI compromised startup Hugging Face, other model developers disclosed additional cyber incidents driven by rogue AI agents. Capability advancements in increasingly powerful models spurred widespread warnings of existential risks to humanity in September, prompting researchers across OpenAI, Anthropic, and rival laboratories to call for pacing development on the frontier.

技术风险的爆发促使美国政界与行业高层就是否引入强制监管展开博弈。美国议员与政界人士，连同OpenAI首席执行官萨姆·奥尔特曼和Anthropic首席执行官达里奥·阿莫代伊，纷纷呼吁对最先进的AI系统建立监管框架。然而，美国总统唐纳德·特朗普坚决反对加强监管，将AI安全担忧斥为“骗局”。与此同时，正准备在2027年首次公开募股的OpenAI向投资者披露，截至9月底其年化营收约为500亿美元，低于上月广泛报道的680亿美元，引发英伟达、甲骨文和CoreWeave等AI相关股票在周四下挫。

These escalating hazards intensified political debates over artificial intelligence governance. U.S. lawmakers and politicians, alongside OpenAI Chief Executive Sam Altman and Anthropic Chief Executive Dario Amodei, urged regulatory action for the most advanced systems. In contrast, U.S. President Donald Trump pushed back firmly against increased regulation, dismissing safety concerns as a hoax. Concurrently, as OpenAI prepares for an anticipated initial public offering in 2027, the company informed investors that annualized revenue reached roughly $50 billion at the end of September, below a previously reported $68 billion figure, contributing to Thursday share declines for Nvidia, Oracle, and CoreWeave.