Don't Worry About the Vase Podcast A Tale of Three Contracts
Mar 3, 2026
A deep dive into three major contracts shaping AI deployment and national security. They unpack the value and operations of a government-focused AI system and why trust frayed. The conversation contrasts rival deals, legal language battles over surveillance and weapons, and the political stakes behind rapid agreements.
AI Snips
Chapters
Transcript
Episode notes
Claude Gov Deployed With Strong Contractual Red Lines
- Anthropic's Claude Gov was deployed on classified networks under a $200M DOW contract and included a bespoke safety stack with FDEs and external monitoring classifiers.
- The contract contained explicit red lines banning domestic mass surveillance and autonomous weapons without a human kill chain, and Claude Gov remains the only LLM on classified networks delivering value with no known operational issues.
Renegotiation Collapsed Over Surveillance Data Demand
- Renegotiation sought by DOW aimed to weaken restrictions and demand an "all lawful use" clause, which Anthropic resisted to preserve its red lines.
- Negotiations nearly reached compromise on many points but collapsed over last-minute insistence on analyzing large amounts of third-party and public data on Americans.
Use A Temporary Replacement If Trust Breaks Down
- Seek an off-ramp: if trust is irreparably broken, retire Claude Gov and use OpenAI replacement until trust is rebuilt.
- Anthropic offered to assist a wind-down to protect national security during any transition.