Iranian-Linked Hackers Used Anthropic's Claude AI to Target U.S. Navy Ships, Company Says

Anthropic disclosed this week that a group linked to Iran used its Claude AI model to help identify and target U.S. naval forces, according to a report the company released and CBS News covered on September 16. The company did not name the group, calling it only a "threat actor." Anthropic said it banned the account and shared its findings with the U.S. government to help disrupt the operation.

According to Anthropic, the group used Claude to compile open-source information, including satellite imagery, ship and aircraft tracking data, and captions on publicly posted military photos, to build a clearer picture of U.S. naval movements and generate targeting recommendations. Anthropic said Claude access is blocked inside Iran, but adversaries can get around that easily using VPNs and foreign phone numbers.

"They're using it to enhance the way they already operate and the tactics or the techniques that they're already used to doing," Nikita Shah, a senior fellow at the Center for Strategic and International Studies, told CBS News. "It just makes it much more effective, much more efficient, much faster."

Sam Bresnick, a research fellow at Georgetown's Center for Security and Emerging Technology, put it more bluntly to CBS News: "It is getting much harder for the U.S. military to cloak or to obscure its activities."

The Anthropic case is not isolated. A July 2026 report from threat intelligence firm Recorded Future found that AI likely accelerated Iran's existing cyber and military capabilities, and that AI-enabled tactics refined in Ukraine's war with Russia may have contributed to Iranian drone attacks against Israel and Gulf states, given Russia's support for Iran's military, CBS News reported.

Caine: U.S. Troops Must Assume They'll Be 'Hunted'

Gen. Dan Caine, chairman of the Joint Chiefs of Staff, delivered a blunt warning on September 16 at the Air Force Association's Air, Space and Cyber Conference, according to Defense Scoop. Citing the Ukraine-Russia war, Caine said the life expectancy of a newly arrived Russian recruit on parts of the front line is now as little as 20 to 30 minutes, driven by cheap AI-guided drones that keep tracking targets even when GPS is jammed or links are cut.

"We have to assume from now on that our formations will be hunted by autonomous systems, jammed across the spectrum, and tracked in real time," Caine told the conference, according to Defense Scoop. He said the joint force, all 2.8 million volunteer members of it, needs those tools "before they need them."

Congress Asks Whether Anyone Is Really in Control

That same week, a different set of warnings surfaced on Capitol Hill. At a September 16 hearing of the Tom Lantos Human Rights Commission titled "Artificial Intelligence in the Military Domain: Implications for Human Rights," witnesses told lawmakers that requiring a human to hit an approval button doesn't guarantee real human control, Military Times reported.

"The approval button alone does not demonstrate the control," said Anna Mysyshyn, a Ukrainian AI-governance researcher, according to Military Times. She said Ukraine's war has already compressed targeting decisions from hours to minutes, leaving operators supervising multiple AI systems without enough time or information to catch a bad recommendation before it's acted on.

Amanda Klasing of Amnesty International USA raised a related concern: AI can compress the military "kill chain" so much it becomes hard to tell whether meaningful human judgment happened at all before a strike, Military Times reported. Joseph Chapa, a former military officer and ethics scholar, argued the debate shouldn't fixate only on how autonomous a system is, but on how it fails and whether its decisions can be audited afterward.

This tension sits directly on top of Pentagon policy. A 2023 Defense Department directive requires autonomous and semi-autonomous weapons to preserve "appropriate levels of human judgment" over the use of force. But the Pentagon's 2026 AI strategy calls for becoming an "AI-first" warfighting force, redesigning workflows around current AI capabilities, according to Military Times. Whether those two goals are compatible as decision speed increases is exactly what the hearing was arguing about, and no resolution has been reached.

A Civilian Death in Zaporizhzhia

In July 2026, a Russian drone struck a gas station in Zaporizhzhia, killing three Ukrainian civilians including a 19-year-old student. According to reporting by The New York Times, cited by Help Net Security, Ukrainian investigators who examined the wreckage found an Nvidia Jetson Orin chip, a commercial computer-vision component not designed for weapons, and concluded the drone's onboard AI selected the specific target without a human making the final call.

Help Net Security's analysis argues that incident marks a shift: earlier AI-assisted drones still relied on remote human piloting, while this one operated independently through the terminal engagement. The piece is careful to say autonomous weapons aren't inherently unlawful and will likely be part of how militaries reduce risk to their own troops going forward. But it argues the Zaporizhzhia strike shows why the international community needs clearer, enforceable rules on when software, rather than a person, gets to make the kill decision. The United Nations has been debating exactly that question, without producing binding rules.

The China Variable

Separate from the Iran and Ukraine cases, some analysts argue the deeper worry is China. James Gorrie, writing in the Epoch Times, points to Beijing's sustained investment in hypersonic missiles, quantum communications, AI-driven command systems, and autonomous drones as capabilities that reinforce each other. AI fuses sensor and satellite data fast enough to feed precision-guided weapons, while quantum key distribution could harden the command networks moving that data. Gorrie is explicit that this is an interpretation of publicly available evidence, not a confirmed statement of Chinese intent, and that the buildup doesn't by itself make conflict more likely.

AI is compressing both sides of warfare at once, making it easier for adversaries to find American forces and harder for human operators, wherever they sit in the chain of command, to catch a mistake before it becomes a body count. Congress has now held a hearing on it. No legislation has been introduced, and the Pentagon has not said how it plans to reconcile a 2023 directive demanding human judgment with a 2026 strategy demanding AI-first speed.