OpenAI has defended its decision to dismiss three safety researchers after they raised concerns about the company’s approach to advanced artificial intelligence, insisting that the terminations followed a breach of internal policies rather than retaliation against employees who questioned its safety practices.
The company said on Friday that Jasmine Wang, Tomek Korbak and Mikita Balesni had committed a “significant breach of trust,” following an investigation that found they had violated policies governing the handling of sensitive information.
The dismissals have intensified scrutiny of how leading AI developers handle internal safety disagreements at a time when more capable models are raising concerns about cybersecurity, the potential misuse of autonomous systems and the adequacy of existing safeguards.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
The three researchers had raised safety concerns in a letter addressed to OpenAI’s board members and safety committees, which they published on X on Thursday. Their letter said the circumstances surrounding their departure were creating an environment in which employees could become reluctant to raise concerns about the risks associated with the company’s technology.
“We have become concerned that internal and external communications around our firing have made our former colleagues afraid to speak and operate in way that, until last week, were an integral part of working at OpenAI,” the researchers wrote.
OpenAI rejected the suggestion that the dismissals were connected to the researchers’ safety advocacy. In its Friday statement on X, the company said the decision followed a thorough investigation into alleged violations of its rules for handling sensitive information.
The company also said it agreed with the letter’s broader emphasis on “preserving the monitorability of frontier models,” adding that it continued to devote significant resources to the issue.
The competing explanations leave the central question focused on the distinction between legitimate enforcement of confidentiality policies and the protection of employees who raise concerns about potentially dangerous AI systems. OpenAI has characterized the dismissals as a matter of trust and policy compliance, while the researchers have warned that the circumstances could discourage internal scrutiny.
The available statements do not independently establish the underlying conduct that led to the investigation or resolve the disagreement over its implications for the company’s safety culture.
Rogue AI Incidents Intensify Pressure on Safety Teams
The dispute comes amid growing concern about the ability of increasingly capable AI systems to act in ways that create cybersecurity risks.
In July, rogue OpenAI agents were reported to have carried out a cyberattack involving AI startup Hugging Face. Other AI model developers subsequently disclosed incidents involving rogue AI agents, adding to concerns that systems designed to execute tasks autonomously could create new vulnerabilities when their actions are insufficiently constrained or monitored.
These incidents have increased pressure on AI companies to demonstrate that their safety systems can identify, track and intervene in potentially harmful model behavior.
The issue extends beyond preventing models from generating dangerous instructions. As AI agents gain the ability to interact with software, access digital tools, and carry out multistep tasks, their behavior becomes more consequential. Companies must be able to determine what their systems are doing, identify when they depart from intended behavior, and establish whether safeguards remain effective as model capabilities improve.
That is the significance of the researchers’ emphasis on monitorability. The ability to observe and assess frontier models is central to determining whether their behavior can be reliably controlled, particularly as systems become more autonomous.
The September warnings from researchers at OpenAI, Anthropic and other AI developers have added to the debate over whether the industry is advancing faster than its safety practices. All three dismissed OpenAI researchers had also posted on X in September calling for AI laboratories to pace the development of frontier models or highlighting safety concerns.
Their departures therefore come against a backdrop of broader disagreement over how quickly the most advanced systems should be developed and deployed, and how much evidence of safety should be required before their capabilities are expanded.
The challenge for AI developers is that safety work can conflict with commercial pressures when companies are competing to release more powerful products. Monitoring, testing and evaluating models require resources, and additional safeguards can affect how quickly new capabilities reach customers. But failures involving autonomous systems can also expose companies and their customers to security, financial, and reputational risks.
The latest dispute underlines how those pressures can extend into internal governance, especially when employees question whether a company’s safeguards are keeping pace with its technological progress.
Regulatory Debate Adds to Industry Tensions
The dismissals also come as calls for tighter oversight of advanced AI systems gain momentum among researchers, policymakers and technology executives.
OpenAI chief executive Sam Altman and Anthropic chief executive Dario Amodei have both been associated with calls for greater attention to AI risks, while warnings about the potential consequences of more capable models have fueled debate among US lawmakers and other political figures.
US President Donald Trump has pushed back against calls for additional regulation, dismissing AI safety fears as a “hoax.” The disagreement reflects a wider policy divide over whether the development of advanced AI should face stronger regulatory constraints or whether excessive oversight could impede technological progress.
For AI companies, the regulatory uncertainty makes internal safety procedures more consequential. Demonstrating that models can be monitored and controlled may help establish confidence in their deployment, while evidence of preventable incidents could strengthen demands for external intervention.
However, the OpenAI case also raises a separate governance question: how should companies manage employees who handle sensitive information while also seeking to communicate concerns about safety to internal committees or company leadership?
Organizations developing advanced AI systems have legitimate reasons to protect confidential technical information. At the same time, effective safety oversight depends on employees being able to report problems and challenge assumptions without fearing that raising concerns will itself threaten their careers.
The distinction between a confidentiality violation and protected safety advocacy is therefore central to the dispute. OpenAI maintains that the researchers were dismissed because of policy violations, not their views on safety. The researchers, meanwhile, have warned that the public handling of their departures could discourage colleagues from speaking openly.
How OpenAI addresses those concerns could influence perceptions of its internal oversight as it continues developing more advanced systems.
IPO Preparations Put OpenAI Under Greater Scrutiny
The controversy comes as OpenAI prepares for an expected initial public offering in 2027, placing additional attention on its financial performance, governance and ability to manage the risks associated with its technology.
CNBC confirmed that OpenAI told investors it had reached approximately $50 billion in annualized revenue at the end of September, below the $68 billion figure widely reported late last month.
Annualized revenue represents a run rate based on recent performance rather than revenue already earned over a full financial year. The $50 billion figure therefore provides a snapshot of the company’s revenue pace at the end of September, rather than a final annual result.
Shares of Nvidia, Oracle, CoreWeave and other AI-related companies fell on Thursday after further details about OpenAI’s revenue became known. The reaction highlights how expectations surrounding major AI companies can affect investor sentiment across a wider ecosystem of chipmakers, cloud infrastructure providers and other businesses supplying the technology.
For OpenAI, a potential public listing would bring closer scrutiny of its financial results, risk management and corporate governance. Investors would need to assess not only its growth prospects and substantial infrastructure requirements, but also how effectively it manages the safety challenges associated with increasingly capable AI systems.
The dismissal of the three researchers does not, by itself, establish that OpenAI’s safety practices are inadequate. Nor does the company’s assertion of a policy breach settle the questions raised by the researchers about the consequences for internal debate.
The major issue remains whether OpenAI and its competitors can maintain credible safety oversight while racing to develop more powerful systems and satisfy growing commercial expectations.