Former OpenAI safety researcher, Leopold Aschenbrenner, has raised significant concerns regarding the security practices at the company, labeling them as “egregiously insufficient.” In a recent video interview with Dwarkesh Patel, Aschenbrenner discussed internal conflicts over priorities, indicating a shift towards rapid growth and AI model deployment at the expense of safety.
Internal Memo and Subsequent Termination
Aschenbrenner revealed he was terminated for documenting his concerns. During a four-hour conversation, he detailed how he wrote an internal memo outlining his safety concerns, which he shared with external experts. Following a significant security incident, he updated the memo and shared it with board members, leading to his swift dismissal.
Firing Questions and Loyalty Issues
Aschenbrenner disclosed the nature of the questions asked during his termination, which focused on his views on AI progress, AGI, the level of security for AGI, government involvement in AGI, loyalty to OpenAI, and his activities during board events. AGI, or artificial general intelligence, refers to AI that matches or exceeds human intelligence across various fields.
Company Loyalty and Employee Solidarity
Loyalty to the company and CEO Sam Altman became crucial after Aschenbrenner’s dismissal. Over 90% of employees signed a letter threatening to resign in support of Altman, popularizing the slogan, “OpenAI is nothing without its people.” Aschenbrenner did not sign this letter despite pressure.
Superalignment Team Dissolution
The superalignment team, led by Ilya Sutskever and Jan Leike, was responsible for long-term AI safety. The departure of key team members and the subsequent dissolution of the team raised further scrutiny. A new safety team, led by CEO Sam Altman, was formed, reporting directly to the OpenAI board.
Contradictions in Safety Priorities
Aschenbrenner pointed out contradictions between OpenAI’s public safety statements and internal actions. He cited instances where security concerns were downplayed when it came to allocating resources or making trade-offs. Leike supported these claims, stating that safety processes were deprioritized under Altman’s leadership.
AGI Development Concerns and Global Competition
Aschenbrenner emphasized the need for a cautious approach to AGI development, especially with fears of China striving to surpass the United States in this area. He warned of China’s efforts to infiltrate American AI labs, which could impact not just technological advancement but the future of liberal democracy.
Controversial Non-Disclosure Agreements
Recent revelations indicated that OpenAI required employees to sign restrictive non-disclosure agreements (NDAs) preventing them from speaking out about safety practices. Aschenbrenner refused to sign such an NDA, despite being offered $1M in equity.
Call for Transparency and Accountability
A collective of nearly a dozen current and former OpenAI employees signed an open letter demanding the right to address company misdeeds without fear of retaliation. The letter, endorsed by industry figures like Yoshua Bengio, Geoffrey Hinton, and Stuart Russell, stressed the importance of transparency and accountability in AI companies.
Sam Altman’s Response
Following the controversy over restrictive employment clauses, Sam Altman claimed unawareness and assured the public that the legal team was addressing the issue. He acknowledged the inclusion of an equity cancellation provision in exit documents, which has since been removed. OpenAI confirmed the release of employees from the non-disparagement agreements and the removal of the contentious clause.
Lack of Comment from OpenAI
OpenAI did not provide a comment to Decrypt regarding these issues.
