AI’s role in helping select battlefield targets has drawn increased scrutiny, after Anthropic PBC’s relationship with the Pentagon deteriorated partly due to concerns about fully autonomous weapons — a nontechnical term meaning arms that can select and destroy targets on their own…
Sankar compared the role of technology in improving the efficiency of targeting to improving other processes like value chain management — a process that often takes months but can be faster.
“The word targeting to the lay public has some sort of connotation, it sounds like someone’s shooting,” Sankar said. “But it’s actually a very bureaucratic, large-scale doctrinal process.”
Chat on Claude:
I just read that Anthropic is being used in the Palantir Maven system. Is that true?
Identified relevant source addressing user’s query directly
Yes, according to the Bloomberg article you shared, that’s correct. Palantir uses Anthropic’s Claude product in its Maven platform.
However, the relationship has hit some significant turbulence. That relationship has been shaken by the Pentagon’s designation of Anthropic as a supply chain risk earlier this month, over the firm’s insistence on safeguards for military use of its AI tools — a move that could force Palantir to seek alternate providers.
Anthropic has contested the Pentagon decision in federal court, with a hearing in the case scheduled for Tuesday.
More broadly, AI’s role in helping select battlefield targets has drawn increased scrutiny, with Anthropic’s relationship with the Pentagon deteriorating partly due to concerns about fully autonomous weapons — arms that can select and destroy targets on their own.
So the short version: yes, Claude is (or was) embedded in Maven, but Anthropic’s insistence on AI safety guardrails for military applications has put that arrangement under real legal and commercial strain.
What AI guardrails does Anthropic put on AI used for weapons systems? Are they public?
Assembled comprehensive overview of Anthropic’s public weapons policy
Yes, they are public — and they’ve been at the centre of a very significant legal and political dispute over the past month or so. Here’s the picture:
Anthropic’s Two Core “Red Lines”
Anthropic publicly stated two specific prohibitions on military use of Claude:
- No fully autonomous weapons — Claude cannot be used in weapon systems that do not need human oversight to launch attacks. On Point Dario Amodei explained the reasoning: frontier AI systems are simply not reliable enough to power fully autonomous weapons, and without proper oversight, they cannot be relied upon to exercise the critical judgment that highly trained, professional troops exhibit every day. Anthropic
- No mass domestic surveillance — Claude cannot be used for mass surveillance of Americans. On Point Anthropic argued that powerful AI makes it possible to assemble scattered, individually innocuous data into a comprehensive picture of any person’s life — automatically and at massive scale — and that fully autonomous weapons and mass surveillance are incompatible with democratic values. Anthropic
What Anthropic Said It Would Allow
Importantly, Anthropic wasn’t opposed to military use broadly. The company said it supports the use of AI for lawful foreign intelligence and counterintelligence missions Anthropic, and offered to work with the Pentagon on R&D to improve reliability of AI systems for defence purposes.
Why It Became a Crisis
The clash was initiated in January 2026 when Defence Secretary Hegseth issued an AI strategy memorandum directing all DoD AI contracts to incorporate standard “any lawful use” language within 180 days — which contradicted Anthropic’s existing contract. ASIS International Anthropic refused, and Hegseth declared the company a “supply chain risk,” restricting military contractors from doing business with it. CBS News
The Court Case
Anthropic sued, and the hearing went well for them. A California judge sharply questioned the government’s position, suggesting the Pentagon may be “attempting to cripple Anthropic” for pushing restrictions on AI weapons, and questioned whether a company can be designated a supply chain risk simply for being “stubborn” and “asking annoying questions.” CBS News
Anthropic’s Full Statement
Their public statement is on their website at anthropic.com/news/statement-department-of-war — it’s worth reading in full if you’re working in this space, as it’s quite candid about both their reasoning and the limits they accept on their own authority.
Given your work on automated intelligence assessment and democratic accountability, Robert, this dispute is directly relevant — it’s essentially a live test case of exactly the institutional accountability questions your Royal Commission submissions addressed.
No mass domestic surveillance — Claude cannot be used for mass surveillance of Americans
Does that mean that Claude can be used for mass surveillance outside of America – like in Australia for instance? Is it being used in this way?
Investigated Anthropic’s international surveillance policies and deployment practices
That’s a sharp and important question. Let me search for what’s actually known.
Identified key finding about foreign surveillance policy distinction
Your instinct here is well-founded, and there’s a direct answer buried in the reporting. The short version: yes, the protection is explicitly limited to Americans, and at least one major publication has flagged it plainly.
The Explicit Gap
Anthropic did not protest the possibility of Claude being used in the lawful mass surveillance of foreign citizens using the same methods. Time That quote is from Time magazine’s deep investigation into the Anthropic/Pentagon dispute, and it’s stated almost as an aside — which makes it more notable, not less.
The framing of Anthropic’s own public statements consistently refers to “mass domestic surveillance” and “surveillance of Americans.” CEO Dario Amodei has said Anthropic will support “national defense in all ways except those which would make us more like our autocratic adversaries.” Scientific American But that framing doesn’t extend to foreign populations.
What Claude Is Actually Doing in the Military
A source directly familiar with the military capabilities of Claude told CBS News the main task Claude is doing is sifting through large amounts of intelligence reports — synthesising patterns, summarising findings, and surfacing relevant information faster than a human analyst could. The targeting process remains human-driven. CBS News
The Grey Zone Problem
The question is whether Claude, operating inside Palantir’s systems on classified networks, could be doing something similar to Israel’s Lavender and Gospel systems — processing intelligence, identifying patterns, surfacing persons of interest — an advancement in autonomous data processing that could transform military intelligence. Scientific American Researchers note that the more capable Claude becomes, the thinner the line between the analytical grunt work Anthropic is willing to support and the surveillance and targeting it has pledged to refuse. Scientific American
What This Means for Australia
There’s no specific reporting on Claude being used for surveillance of Australians or other Five Eyes allies’ populations. But the structural gap you’ve identified is real:
- Anthropic’s contractual red lines protect Americans specifically
- Foreign nationals — including Australians — appear to have no equivalent contractual protection
- The US and Australia share intelligence through Five Eyes, meaning data processed by AI tools on classified US networks could plausibly include Australian signals intelligence
This is precisely the kind of gap your Royal Commission submission work touches on — where the absence of legislative guardrails leaves protection dependent on the goodwill of a private company, and that company’s goodwill is itself geographically bounded.