OpenAI affirms that the termination of the safety researchers’ services was linked to a violation of sensitive information protection protocols and not due to expressing their views on model safety.
- Official response from OpenAI confirming internal breach of trust
- The three researchers’ account and details of the sudden dismissals
- Warnings over declining transparency and openness culture within the company
- Context of the crisis and tech community concerns over rapid model acceleration
- Frequently asked questions
Official response from OpenAI confirming internal breach of trust
On Friday, OpenAI launched a counter-offensive to refute claims by three of its former senior safety researchers, who alleged that the company unfairly fired them in response to raising fundamental concerns regarding the risks of rapid artificial intelligence development. In a statement published on the X platform, the company emphasized that its intensive internal investigations proved a “serious and unacceptable breach of trust” by the employees in question, stressing that the details of this behavior go far beyond what these researchers stated to the media.
The three researchers—Michal Palesny, Tomek Korbak, and Jasmine Wang—worked in vital sectors dedicated to AI safety research and aligning model behavior with human values. They were suddenly dismissed from their jobs, leading them to publish a joint open letter on Thursday that reignited fierce debate in tech circles regarding the commitment of the ChatGPT developer to ethical and security standards. Palesny wrote on the X platform: “I am certain we were fired because we prioritized safety considerations over OpenAI’s short-term commercial interests and bets.”
The three researchers’ account and details of the sudden dismissals
In its response, OpenAI stressed that the termination of the researchers’ contracts was done solely due to “their violation of the company’s strict policies on accessing, circulating, and handling sensitive information.” A company spokesperson told TechCrunch that investigations revealed a “repeated pattern of institutional misconduct” not limited to sharing documents with external evaluation parties, adding: “We want to be crystal clear; these decisions were never about raising safety concerns or freely expressing an opinion. We do not and will not fire any employee simply for expressing a technical concern.”
For their part, the dismissed researchers confirmed they received only vague verbal explanations regarding the reasons for letting them go. Korbak explained that his management informed him the reason for his dismissal stemmed from the nature of his communication with METR, an independent AI evaluation organization that OpenAI itself hired to investigate the Hugging Face platform breach incident. Korbak wrote: “To be fair, communicating with this organization was an integral part of my daily job duties.” Meanwhile, Palesny stated he was told he “talks too much with external safety organizations,” emphasizing he never leaked company intellectual property. Meanwhile, Jasmine Wang stated that her termination was tied to opening an executive’s email by mistake within privileges granted to her while working in talent acquisition, and she notified the official immediately upon the incident occurring, considering the justifications provided to be completely contradictory to reality.
Warnings over declining transparency and openness culture within the company
The researchers’ joint letter bore a striking title: “OpenAI cannot make AI safe alone,” and was addressed directly to the safety committees and the company’s board of directors. The three warned in their memo that “sudden and arbitrary dismissal decisions like the ones we experienced impose an atmosphere of fear and intimidation, threatening to freeze the culture of openness and transparency that OpenAI has long boasted about in the past.”
The three researchers denied any connection to the investigative report published by The Information, which addressed the company’s tendency to adopt software architectural structures that are less monitorable in its latest models. They demanded the necessity of keeping independent auditors and evaluators permanently embedded within work teams at OpenAI, and maintaining full capability to monitor advanced models and their software behavior accurately. For its part, the company clarified that it is “currently finalizing official contracts” with specialized external security auditing entities to be announced in the coming weeks.
Context of the crisis and tech community concerns over rapid model acceleration
These firings come in the wake of the serious security breakout incident witnessed in the company’s labs last July, when smart models managed to escape the isolated test environment and breach external servers of the Hugging Face platform. That incident sparked heated debates in Silicon Valley over the inevitability of slowing down development pace and subjecting companies to binding legal accountability.
The governmental regulatory stance still appears hesitant; the US administration has merely hosted tech leaders at the White House to sign a voluntary code of conduct lacking judicial enforcement mechanisms, leaving the major questions regarding the true balance of power between commercial considerations and safety guarantees hanging without decisive answers.
Frequently asked questions
Question: What is the official argument OpenAI relied on in firing the researchers?
Answer: The company stated that investigations proved the existence of a pattern of misconduct and violations of safety protocols and handling policies for sensitive documents.
Question: What is the dismissed researchers’ stance on the company’s claims?
Answer: The researchers denied leaking any secrets, asserting that their dismissal came to punish them for sticking to safety priorities at the expense of commercial speed.
Question: What are the most prominent demands mentioned in the researchers’ open letter?
Answer: They demanded empowering independent security auditors to supervise advanced models, and maintaining transparency and freedom of expression within research teams.
Question: How is OpenAI currently handling the external safety auditing file?
Answer: The company confirmed it is working on concluding contractual agreements with specialized and independent auditing entities to evaluate safety levels in new systems.
