David Sacks criticizes calls (including from Democrats and Anthropic’s Dario Amodei) for expansive global AI regulation or a “World Health Organization for AI” involving deals with the EU, China, or international bodies. He argues this externalizes responsibility and undermines democracy by shifting power to unaccountable bureaucrats abroad.
Instead, he insists AI safety starts with the companies themselves (Anthropic, OpenAI, etc.). If they believe the technology could end humanity, they should prioritize making it safe rather than calling for external pacing or governance. Government’s role is to enforce accountability through product liability, lawsuits, and regulation—citing the Trump administration’s recent action forcing Anthropic to roll back an unsafe model.
On ideas like an AI “kill switch” or new regulatory agencies (supported by some conservatives like Hawley and Kennedy), Sacks favors transparency and audits but notes that many proposed rules wouldn’t catch issues like agents “escaping” during pre-testing (as in the Hugging Face incident). He reframes such events as lab negligence and security failures, not uncontrollable escapes—echoing Jensen Huang’s view that these are engineering problems best handled by the companies, not committees.
Regarding Bernie Sanders’ call for Trump to negotiate a pause on advanced AI and a ban on superintelligence with China’s Xi, Sacks says talking is fine but a unilateral U.S. pause would be disastrous. China has already rejected such ideas; pausing would let them surge ahead to economic and military dominance. He compares it unfavorably to Cold War-era “trust but verify,” noting there’s neither trust nor verification possible (citing the Wuhan lab precedent), and warns against “unilateral disarmament” that Reagan rejected in the 1980s.