Anthropic built its brand on AI safety. Now its flagship product is helping the U.S. military plan air strikes on Iran.
According to sources familiar with the matter, Claude AI systems have become a key tool in military planning operations targeting Iran, even as Anthropic and the Defense Department have butted heads on the terms of that relationship. The revelation has prompted lawmakers to push for greater oversight of how AI tools are being deployed in active military contexts.
The Safety Company Goes to War
Anthropic has long positioned itself as the responsible counterweight to OpenAI and other labs racing to deploy AI as fast as possible. The company's Responsible Scaling Policy, its emphasis on constitutional AI, and its public communications all center on careful, measured deployment. Military strike planning sits uncomfortably alongside that image.
The tension is not new. Anthropic updated its usage policy in late 2024 to permit certain national security applications, a shift from its earlier, more restrictive stance. But there is a wide gap between "national security applications" as a policy category and "helping plan bombing runs" as a concrete use case. The public is now seeing what that gap looks like in practice.
What Congress Wants
Lawmakers are calling for oversight mechanisms, though the details of what that looks like remain unclear. The core questions are straightforward: What role does AI play in target selection? What decisions remain fully human? And who is accountable when an AI-assisted strike goes wrong?
These are not hypothetical concerns. AI systems can process satellite imagery, logistics data, and intelligence reports faster than any human team. That speed is the whole point of using them. But speed in military planning compresses the window for human judgment, and compressed judgment windows in strike planning have historically led to civilian casualties.
Where Anthropic Stands
Anthropic has not publicly commented on the specifics of its military contracts. The company's clashes with the Defense Department suggest the relationship is not frictionless, but the fact that Claude is actively being used indicates those disagreements have not been dealbreakers.
For the broader AI industry, this sets a precedent. If the company most associated with AI safety is providing tools for military strike planning, it becomes harder for any AI lab to draw a clean line between commercial and defense applications. Google famously faced internal revolt over Project Maven in 2018. Eight years later, the industry's relationship with the military has quietly normalized.
For everyday users of Claude, nothing changes about the product. But it does change what you are implicitly supporting when you pay for a subscription. That is a calculation each user will have to make for themselves.