Interview, Fireside Chat
Anthropic’s CEO explains why he took on the Pentagon
- Current technology lacks readiness for full autonomy, with aircraft-like systems restricted by specific altitude and turning speed limits to ensure safety.
- A substantially limited contract, active potentially until the present, facilitated restricted use cases in Venezuela and Iran without ground-level issues, while future governance discussions focus on forward-looking operational requirements.
- The Department of War expects full operational capability from suppliers, viewing them through a fighter jet analogy where tools must remain usable in combat without provider restrictions, leading to an agreement on approximately 99% of contract terms while broader initiatives remain in early stages.
- The logical trajectory of total Pentagon control, where suppliers cannot dictate service usage, risks the effective nationalization of AI companies working with the defense sector.
- Long-term national security concerns regarding AI span beyond the current administration to future Congresses and geopolitical shifts, specifically addressing the potential for models to become security risks or for individuals to use the technology for destruction.
- Autonomous systems face objections centered on reliability and oversight, creating a dilemma where improper handling could allow a small group or single individual to command vast numbers of drones, such as 10 million.
- Future years will require ongoing discussions to balance the risk of private actors exceeding government power against the danger of the government becoming so powerful it cannot be checked, a tension applicable to both democratic and autocratic regimes.