Overview
- Anthropic released a policy paper forecasting that artificial general intelligence could be achievable by 2028 and outlining two futures in which either democracies set AI rules or authoritarian regimes drive automated repression.
- To slow capability leakage, the paper urges tougher export controls on advanced chips, blocking covert data center access, and defenses against distillation attacks, which train smaller models on the outputs of more powerful systems.
- The company argues that keeping a 12 to 24 month edge in high-end compute and model capability would let the U.S. shape global standards for transparency and safety.
- Anthropic warns that more capable systems could enable mass surveillance, political control, and offensive cyberattacks, and it flags potential military uses by China’s People’s Liberation Army.
- The push drew immediate criticism from some analysts as self-serving, landed as President Trump met Xi Jinping in Beijing, and followed weeks of scrutiny over Anthropic’s restricted Mythos model and U.S. military AI agreements with other major tech firms on May 1.