Three former OpenAI researchers said they were abruptly fired last week for their interactions with outside AI safety organizations. They warned that such decisions could make employees afraid to openly discuss the risks of artificial intelligence. OpenAI itself confirmed its decision, but explained it differently.
Image source: Mariia Berezovsky / unsplash.com
In a letter titled “OpenAI will not be able to ensure the security of artificial intelligence alone,” former laboratory employees noted that this “unusual company”and a key part of its work is to work with third parties to resolve security issues, and its employees must do this “free and without fear”.
Letter addressed to OpenAI Security Committees published Mikita Balesni is one of the fired employees who worked on AI goal alignment issues. The document was also signed by fired Tomek Korbak and Jasmine Wang, who worked in the security field.
On their personal pages on social networks, they wrote that the reasons for the dismissal were explained to them only verbally, and their actions did not go beyond the scope of their official duties. “The reasons for our dismissal simply do not add up to a logical picture.”— wrote Wang. Balesny was informed that the company no longer trusts him because he “communicated too much with third party security organizations”. “I have never disclosed information that constitutes the company’s intellectual property. My work was coordinated with immediate management, heads of research areas and the board of directors”— wrote He.
Korbak was verbally explained that his dismissal was due to “the nature of communication with METR” is a third-party, non-profit research organization that evaluates AI models and security risks. “Let me be clear: communication with METR was part of my job responsibilities.”— added He. Former colleagues of the researchers will now “being afraid to speak up and act on what, just last week, was considered an integral part of working at OpenAI. Layoffs like this—sudden and carried out in such an abrupt manner—undermine the open culture that OpenAI has valued in the past.”the letter says.
The researchers gave OpenAI three pieces of advice: bring third-party security auditors into the organization, ensure it can monitor advanced AI models, and support “a culture of open and transparent dialogue between security researchers and the rest of the security ecosystem”.
In a memo sent to employees, one of OpenAI’s research leaders wrote that he fully supported these recommendations. “We greatly valued their contributions to AI safety, as well as their willingness to speak up and challenge existing ideas. We supported their initiatives and placed great trust in them. I want to emphasize that the reason for these decisions was not that they raised security issues or spoke out openly. We have always encouraged and will continue to encourage such activity. We do not fire employees for highlighting problems.”says the document.
However, the decision to fire these three employees was made “following a thorough investigation which revealed that they violated strict rules regarding the handling of confidential information”the company emphasized on the social network X. OpenAI shares the principle of necessity set out in the letter “ensure the ability to control advanced models”as well as the ideas expressed in the letter regarding “preserving the ability to monitor advanced models”and the company continues to invest significant resources in this area.
If you notice an error, select it with the mouse and press CTRL+ENTER.
