David Sacks, who chairs the President's Council of Advisors on Science and Technology and served as the White House's AI and crypto czar until March, has urged Anthropic and OpenAI to pursue their research ambitions while ceasing demands for special legal treatment. Rather than seeking antitrust waivers or formal approval regimes, Sacks contends that existing product liability frameworks and competitive market dynamics already constrain their behaviour adequately.
In his message to the two companies' leadership, Sacks stated: "The easiest way not to build superintelligence is for you to agree not to build it." However, he rejected the accompanying requests attached to their proposals. He accused the firms of disguising cartel formation as necessary policy, writing that they should abandon the pretence that antitrust law must be suspended "so you can form a cartel", and cease demanding regulatory approval processes that would circumvent product liability frameworks.
Sacks also raised concerns about the independence of METR, the evaluation organisation that Anthropic relies upon, noting that it maintains entangled relationships with the company's investors and personnel.
Liability as discipline
Sacks proposed an alternative enforcement mechanism centred on product liability. Should a model facilitate a significant cyberattack, he argued, the developer would face legal claims, and market mechanisms already penalise systems exhibiting erratic behaviour. Conditioning restraint on regulatory concessions, he suggested, would constitute "blackmail".
His position arrives amid internal dissent at Anthropic. A pretraining researcher departed the company on 8 September, citing concerns that the sector was endangering lives. The company's alignment science lead has assessed the probability of AI causing human extinction within a decade at above 10%.
Europe's approach
The European Union reached comparable conclusions regarding liability frameworks. In February 2025, the Commission abandoned its proposed AI Liability Directive, determining that revised product liability regulations already encompassed AI software adequately. Beginning 9 December, these regulations classify standalone software, operating systems, and AI systems as products subject to strict liability, eliminating the requirement for claimants to demonstrate negligence.
Under the updated framework, defectiveness encompasses cybersecurity weaknesses and self-learning capabilities that become unsafe following deployment. Courts possess authority to compel disclosure of evidence and may presume defectiveness when claimants encounter excessive obstacles in establishing one.
Dual regulation in practice
Europe has not, however, selected a single regulatory path. The AI Act's systemic risk obligations have applied to frontier models since August 2025, and since 2 August the Commission can mandate evaluations and impose fines reaching 3% of global revenue. Consequently, both the approval-style regime that Sacks opposes and the liability-centred approach he advocates will govern the same models simultaneously.
European legal professionals have been analysing the implications of this dual framework since summer, examining applications ranging from driver assistance systems to conversational AI. Sacks is essentially proposing an experiment that one jurisdiction is preparing to conduct.
Source: The Next Web
