Sankar on Face the Nation: Department of War dropping Anthropic reinforces the model-neutral orchestration thesis…
The Department of War has stopped using Anthropic's products after Anthropic insisted on safety guardrails for military use (refusing to remove its restrictions on autonomous weapons and mass surveillance).
Read more
Sankar said publicly that the switch had no operational impact, which gives Palantir's model-neutral orchestration-layer moat a first-hand, government-level data point. Which underlying large model wins or loses does not decide Palantir's fate: Palantir sells the orchestration and ontology layer that gives institutions the ability to switch models and control the weights. In the same interview Sankar addressed the strike on an Iranian school that killed 123 children, saying the likely cause was human error. The ethics tail risk for military AI deepens in step.
Key evidence
- In February President Trump ordered a six-month phase-out of Anthropic over its insistence on safety guardrails for military use, and Hegseth's supply-chain-risk designation echoed it. No stop date is on the record: the only on-record statement is a Pentagon official telling the BBC on Monday 5 Oct that it 'has ceased the use of Anthropic products'
- Sankar said publicly on Face the Nation that the switch had no impact on operations
- Sankar's core argument: an institution's sovereignty and agency means deciding which model to use in each scenario, having 'the ability to switch models' and controlling the weights
- First-hand, government-level confirmation that the orchestration layer is stickier than any particular model (not the sell-side analyst line)
- Same interview: on the mistaken strike on an Iranian school that killed 123 children, Sankar called Bloomberg's report of AI over-reliance inaccurate, said the likely cause was human error, said final authorisation for a lethal strike rests with human commanders, and said to wait for the final report
- Adds to the UK governance controversy: the NHS contract, the lawsuit over the £50 million Met Police contract, and Sankar's 'mob rule' remarks
Catalysts to watch
- Whether later government contracts keep model neutrality as the procurement line
- How frontier labs trade off guardrails for military use (who is dropped after Anthropic)
- Renewal timing of UK and European government contracts (whether the governance backlash affects commercial business)
Risks
- The deeper military AI penetrates, the larger the exposure to reputational and political backlash from a single battlefield incident
- The NHS, the Met Police and the 'mob rule' remarks add up to a running governance tail risk in the UK
- The political and ethical backlash could affect government and commercial renewals. It is not a fundamentals variable, but it needs continuous monitoring
- Noise such as rating upgrades, record highs and price targets is excluded. The Armada modular data centre and UK contract renewal items were already covered on 11 Oct
Sources (1)
- CBS Face the Nation (Sankar interview, aired Sunday 11 Oct)
A brief is a lead for our research, not a verified conclusion or advice.