What happened
A proposed 2024 federal regulation that would have required AI companies to report on their frontier model experiments and safety protocols was removed from the U.S. government’s rulemaking docket. According to The Lever, this withdrawal occurred after major AI firms lobbied against the requirements, citing potential costs and administrative burdens, and following a $25 million donation to a pro-Trump super PAC by an OpenAI executive and his spouse.
In September 2024, federal officials initiated a rule requiring AI companies to disclose details regarding experiments conducted at the technological frontier. The stated goal was to mitigate risks of dangerous accidents and potential loss of life.
The Lever reports that this initiative was subsequently dropped from the government's rulemaking docket. Documents reviewed by the outlet indicate that AI companies lobbied against the rule, arguing that the reporting requirements were overly burdensome and costly.
The report highlights that the withdrawal of the rule followed a $25 million contribution to the MAGA, Inc. super PAC by an OpenAI executive and his wife. Currently, there is no general federal mandate requiring companies to routinely report on the security, development, or unexpected behaviors of frontier models.
In the wake of recent high-profile hacking incidents involving OpenAI and Anthropic, the lack of standardized reporting has left regulators without the instrumentation needed to verify the nature of AI-related threats.
Source details: levernews.com ↗
Why it matters
The absence of this reporting requirement leaves federal regulators and national security officials with limited visibility into the development and security infrastructure of cutting-edge AI models. Experts cited by The Lever suggest that mandatory disclosure would have forced more deliberate development practices and provided authorities with the necessary data to assess real-world risks, such as recent cybersecurity incidents involving major AI firms, rather than relying on industry self-regulation.
The lack of federal oversight creates a 'blind spot' for national security officials. According to Vishal Misra of Columbia University, the current environment is 'much looser' than what was proposed, preventing experts from distinguishing between genuine security threats and public 'doomsday' scares.
Mandatory reporting acts as a deterrent against reckless development. Calli Schroeder of the Electronic Privacy Information Center noted that if companies know they must justify their development processes to an authority, they are more likely to address security gaps and avoid sloppy practices.
The report contrasts the current industry stance—where companies publicly claim to support regulation—with their private lobbying efforts to weaken or eliminate specific disclosure requirements, highlighting a significant gap between public rhetoric and private policy influence.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?
What to watch next
Legislative efforts are currently underway to establish a federal agency for AI oversight. Lawmakers, including Sen. Bernie Sanders and Sen. Peter Welch, have introduced bills that would mandate the types of disclosures previously proposed in the 2024 rule. The effectiveness of these legislative attempts to restore federal oversight remains a key point of contention as the industry continues to face public scrutiny over safety and security.
New legislative proposals, such as those co-authored by Sen. Peter Welch, aim to create an independent expert agency to oversee and mandate incident reporting.
The debate over whether the U.S. government can effectively regulate AI without industry cooperation remains central to upcoming policy discussions.
Observers will be monitoring whether these new legislative bills can overcome the same lobbying pressures that led to the abandonment of the 2024 reporting rule.