← All episodes

FULL INTERVIEW: Ben Thompson on Anthropic v. DoW

| 2 products mentioned
TBPN TBPN host
Ben Thompson guest
Watch on YouTube ai governance geopolitics anthropic government regulation surveillance law semiconductor policy private vs. public power

Ben Thompson discusses his controversial article on Anthropic vs. the Department of War, arguing that AI companies operating as private entities will inevitably face government pressure as AI becomes a source of geopolitical power. Thompson contends that this conflict is not primarily about surveillance concerns but reflects a fundamental tension between private control of transformative technology and state sovereignty, drawing parallels to nuclear weapons regulation and historical precedents like Intel's relationship with government contracts.

Key takeaways
  • AI alignment discussions must grapple with geopolitical realities and the fact that governments will assert control over technologies they view as sources of national power, not just abstract ethics about humanity broadly.
  • The surveillance argument against military AI use is Thompson's strongest critique of Anthropic's position, as current laws were designed assuming physical friction that AI eliminates, making new legislation necessary rather than unilateral corporate restrictions.
  • Private companies cannot sustainably build AI without selling to the commercial and government markets simultaneously—the capital costs (hundreds of billions annually) require mass-market adoption, fundamentally different from how the government built nuclear weapons.
  • Thompson advocates for Congress to pass new digital surveillance laws rather than having private executives make unilateral decisions, warning that relying on corporate moral judgment represents a dangerous abandonment of democratic governance.
  • The Taiwan-China dynamic exemplifies the core problem: restricting chip sales to China creates incentives for military action, while allowing sales creates shared dependence—illustrating that all geopolitical solutions involve intolerable tradeoffs.
  • OpenAI's approach of agreeing to Pentagon oversight while reserving the right to block certain model capabilities effectively creates a "jailbreak competition" with the government, positioning it more favorably with the public than with the tech industry's talent base.

Mentioned (2)

TensorFlow
TensorFlow "What they were actually exposing to the government was TensorFlow APIs that would run on Google h..." ▶ 30:13
The Making of the Atomic Bomb "Dario Amade is a fan of this book, The Making of the Atomic Bomb" ▶ 12:52