# Holding AI Companies Accountable as a Professional Career Path Internal and external advocacy is essential to mitigate catastrophic risks in frontier AI development. - **Source:** [80,000 Hours](https://80000hours.org) - **Published:** 2026-11-03 - **Topic:** [Leadership](https://brifing.si/c/leadership) - **Original article:** - **On Brifing:** https://brifing.si/s/32181546-d2a2-42a3-8b94-10ac3f26df77 ## The Rise of AI Accountability Holding major AI firms accountable has become a critical, underrated career path. Recent scandals at Anthropic, OpenAI, and Google DeepMind highlight the urgent need for oversight. As development speeds increase, waiting for government legislation is often insufficient, making immediate public and expert pressure a vital mechanism for ensuring corporate responsible behavior. ## Anthropic's Controversial Federal Bill Anthropic recently pushed a federal AI bill that would significantly hinder the government's ability to stop dangerous model training. The proposal requires the Commerce Secretary to prove imminent catastrophic risk in a DC court within 45 days. This high legal threshold creates a dangerous bottleneck for preventing potential global AI-driven disasters. ## Preemption of State Regulations The proposed federal legislation would retroactively invalidate flagship AI safety laws in California, Illinois, and New York. By stripping states of their power to halt models posing imminent risks, the bill centralizes control. This shift leaves states unable to protect citizens, relying entirely on a single, potentially slow-moving federal judicial process. ## The Risk of Centralized Power Anthropic claims to fear political abuse of power, yet their bill centralizes control within the federal executive branch. This creates a single point of failure. If the federal government is captured or compromised, there are no state-level safeguards remaining to intervene, effectively leaving corporations to self-regulate their most dangerous research activities. ## Questionable Legislative Motivations Anthropic's push for federal preemption is perplexing given their CEO's public warnings about 25% risks of catastrophic AI outcomes. By limiting state oversight, the company makes it harder to stop reckless competitors. Experts argue that allowing state attorneys general to petition courts would provide a necessary, more robust safety check. ## OpenAI's Safety Researcher Firings OpenAI recently fired three safety researchers, Tomek Korbak, Mikita Balesni, and Jasmine Wang, in simultaneous meetings. While the company cited a vague breach of trust, reports suggest the firings stemmed from a leak investigation regarding 'looped transformers.' The lack of concrete evidence has fueled concerns about how OpenAI handles internal dissent. ## The Leak Investigation Context The investigation focused on whether researchers leaked information to outside researcher Ryan Greenblatt. Because Korbak and Balesni used disappearing messages in their chats, they could not prove their innocence regarding the leak. OpenAI appears to have terminated them based on suspicion alone, despite all parties involved strongly denying any unauthorized disclosures. ## Internal Dissent and Accountability The firing of researchers who advocated for monitorable AI chain-of-thought processes suggests a culture that discourages internal questioning. When staff members who prioritize safety are removed, the company loses critical checks against reckless development. This environment makes it increasingly difficult for employees to voice substantive concerns about dangerous research practices. ## Google DeepMind's Gag Clauses Reports revealed that Google DeepMind utilizes lifetime gag clauses in staff contracts. These agreements prevent employees from criticizing company actions indefinitely. Such restrictive policies stifle transparency and prevent the public from understanding the risks associated with advanced model development, further necessitating external investigative efforts to uncover critical information about operations. ## Leveraging Talent Competition AI companies are highly sensitive to their public image because they compete fiercely for top-tier talent. When staff speak up or threaten to leave, companies are forced to reconsider their ethical stances. This dynamic makes internal advocacy a powerful tool for change, as firms fear being perceived as the 'bad guys'. ## The Role of External Advocates Outsiders, including investigative journalists and policy advocates, play a crucial role in holding AI firms accountable. By digging for information and explaining complex risks to the public and politicians, they create the pressure necessary for reform. This work is highly effective, even with modest resources, due to the companies' vulnerability. ## Future of AI Oversight As AI capabilities advance, the capacity to hold firms accountable is becoming a defining challenge for the industry. Whether through internal whistleblowing or external public pressure, maintaining this oversight is essential. Professionals must continue to push for transparency to ensure that the development of frontier models does not outpace our safety. ## Sources covering this story - https://80000hours.org/podcast/episodes/hold-frontier-ai-companies-accountable/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._