Why this needs care

The phrase 'Claude is used in targeting systems' is doing a lot of work, and it can mean wildly different things — from 'a model helps an analyst sift intelligence' to 'an AI decides who dies.' Those are not the same claim, and the accuracy of the second is far weaker than the first. Here is what the record actually supports.

What is confirmed

In November 2024, Anthropic and Palantir announced a partnership, with AWS, to make Claude models available to US intelligence and defense agencies. Palantir stated it would be the first industry partner to bring Claude into classified environments, operationalised inside Palantir's AI Platform (AIP). Palantir described the use cases as processing large volumes of complex data, identifying patterns, streamlining document review and helping officials make 'more informed decisions in time-sensitive situations.' This is on the record from Palantir's own investor release and Business Wire.

Separately, Anthropic was awarded a DoD prototype agreement worth up to $200M in 2025 (since terminated — see our Anthropic–Pentagon piece).

What is reported (credibly, but as reporting)

Scientific American reports that Claude has become a component of Palantir's Maven intelligence-analysis program, which the US military uses to help identify potential targets. NBC News separately reported the military using AI to help plan air operations, with lawmakers calling for oversight. In these accounts, AI generates or surfaces candidate targets; a human operator approves before any strike. That is meaningfully different from an autonomous weapon.

What is NOT substantiated

There is no credible sourcing for the strongest version of the claim — that Claude autonomously selects targets and authorises lethal force without a human. We are not asserting that, because no reliable source establishes it, and Anthropic's usage policy explicitly prohibits lethal autonomous warfare without human oversight.

The honest grey zone

The legitimate debate is narrower and harder. Anthropic's red lines target fully autonomous weapons and mass surveillance. They do not address a subtler problem critics raise: an AI embedded in a kill chain where a human is nominally 'in the loop,' but where the system surfaces targets at a pace and scale — critics cite the reported thousand-targets-in-24-hours tempo of Israel's Lavender system as the archetype — that can make human review more ceremonial than substantive (Lawfare; Scientific American).

So the precise, defensible statement is this: Claude is confirmed to be deployed in classified US defense and intelligence settings, and is reported to feed analysis into a targeting program where humans approve strikes. It is not established that Claude fires weapons or selects targets autonomously, and Anthropic's policy forbids exactly that. Anyone asserting more than this is going beyond the evidence.