Anthropic CEO Dario Amodei is explicitly calling for democratic nations to form a bloc that chokes off China's access to chips and semiconductor equipment. In a policy essay published June 10, he warns that within one to two years, as superintelligent AI emerges, countries in possession of advanced AI will hold a military edge over those without it akin to "World War II Marines versus medieval knights."
To secure a decisive advantage for democratic allies, Amodei proposes a joint alliance that freely shares chips and semiconductor equipment among members while tightening export controls against adversaries. To prevent superintelligence from becoming a surveillance and power-grabbing tool for authoritarian regimes, he argues democracies should ban fully autonomous weapons in law enforcement and close loopholes that allow mass surveillance through commercial data brokers.
On safety governance, large models must move from voluntary transparency disclosures to a mandatory testing and auditing regime similar to the Federal Aviation Administration. Models exceeding a certain compute threshold would need third-party safety assessments before release, testing for bioweapons synthesis, cyberattacks, and autonomous loss of control. If an evaluation indicates a model poses a public threat, the government should be authorized to block its release. Developers would also need to protect model weights at defense-grade levels to defend against state-sponsored hackers.
Addressing the challenge of mass unemployment, the essay argues that AI will gradually replace human cognitive work, with the core bottleneck being distribution of gains rather than growth. Anthropic co-published a job security policy framework that proposes a combination of wage insurance, capital gains taxes, and universal basic income — and pledged funding to support it. In healthcare, the company suggests using AI simulations to shorten drug approval processes and accelerate the delivery of new clinical treatments.