On Point with Meghna Chakrabarti Why the Pentagon wants AI without guardrails
13 snips
Mar 2, 2026 
Heather Roth, a DOD AI ethics researcher who helped write the Pentagon’s 2020 AI principles, and Stephen Levy, longtime Wired technology journalist, discuss the Anthropic–Pentagon dispute. They explore how the Pentagon uses generative AI in strikes and surveillance. They debate safety-first limits, supply-chain designations, legal lines around domestic surveillance, and where autonomy in weapons becomes dangerous.
AI Snips
Chapters
Transcript
Episode notes
Anthropic's Red Lines Limit Military Use
- Anthropic refused Pentagon demands to allow its Claude AI for mass domestic surveillance and fully autonomous weapon systems.
- CEO Dario Amodei framed this as defending democratic values while supporting military use with strict guardrails against misuse.
Supply Chain Label Threatens Anthropic's Survival
- The Pentagon's supply chain risk designation could effectively bar Anthropic from selling to many government-linked customers.
- Stephen Levy notes the designation risks killing Anthropic's business model that targets corporate and government contracts.
Palantir Claim Sparked The Confrontation
- Anthropic believed its contractual safeguards would prevent military misuse but a disputed Palantir report about a Venezuela raid triggered concern.
- Palantir allegedly told the Pentagon Claude was used in that raid, prompting the Pentagon to push for removing Anthropic's red lines.