Anthropic PBC says people tried to bend its Claude AI toward warfighting and spying. In a threat report published Thursday, the company outlines efforts touching kamikaze drone swarms, missile navigation and projects connected to potential biological weapons.
The report identifies users in Iran, Russia, China and Yemen among those flagged over the past year. Anthropic also says the incidents did not involve its top-end models, Fable and Mythos.
"As models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer," the company wrote. It adds that US AI players are racing to build stronger systems, in part to stay in front of Chinese rivals, while trying to keep safeguards in place. Anthropic says some attempts aimed to aid America's enemies and tried to slip past Claude's protections or hide user locations.
Anthropic details five examples tied to work that could advance biological weapons. One case involved state-sponsored virologists seeking an enhanced strain of chikungunya, a mosquito-borne virus that can debilitate people. The activity surfaced when Claude was asked to draft a grant proposal tied to a military institute located in an unnamed nation where Anthropic does not operate. The company says all the bio examples were ambiguous and could reflect legitimate science.
On the battlefield side, Anthropic says non-state-affiliated users in Russia tried to build software to run a swarm of FPV loitering munitions that would learn from Ukrainian combat footage.
In Houthi-controlled northern Yemen, operators turned to Claude to develop programs for several different missile types. "We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket," Anthropic said, adding that the users returned to Claude hours later to troubleshoot why it failed.
Separate operations connected to China and Iran aimed to build and enable surveillance. One China-linked effort used Claude to "track, profile and recruit Uyghurs" in the Syrian Army. An Iranian unit built surveillance code disguised as a prayer times app. China's foreign ministry did not immediately reply to a request for comment.
In fast changing tech moments, steady investing habits help protect and grow savings. Join Briefs Finance CEO Jaspreet Singh on September 29th for a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, where he shows how we're spotting investment opportunities as the dollar falls. Save your spot.
The system could assemble "intelligence dossiers" for any specified number without a court order. Anthropic says it banned the account, but the Claude-powered platform had already been installed locally and was no longer under the company's full control. A request for comment did not receive an immediate reply from the Malian government.
The report highlights how hackers are leaning on AI to move faster. One group that Microsoft Corp. calls Midnight Blizzard and links to Russia's intelligence services "increased their speed by automating their operations using AI," including using Anthropic's software to automate many phases of malware development and to craft phishing emails targeting military intelligence and defense organizations. A request for comment to Russia's foreign ministry did not receive an immediate response.
Anthropic says the ShinyHunters cyber-extortion outfit used AI to scan for leaked credentials and then used those to steal data from affected organizations. The company notes it is not disclosing any new or unusually dangerous weaknesses in critical systems. "The cases we share here aren't typical misuse, but rather examples of the most notable and novel threat activity we've identified to date," it wrote.
Anthropic's stance has strained ties with the Pentagon, which suspended collaboration once the company refused to greenlight Claude's use for expansive spying or independently operated weapons.
The report lands as safety worries get louder. On Tuesday, Researcher Jacob Coxon left Anthropic, saying he was worried that the company and OpenAI are "gambling with our lives," and adding that teams building AI "earnestly believe it could kill us all by the end of the decade." In response, current Anthropic employee Evan Hubinger said that he and colleagues do worry about that scenario.
For investors, here is the tension to watch: demand for cutting-edge AI keeps climbing, with US firms pushing forward partly because of competition with China, while mounting safety, legal and national security concerns can slow deals, reshape contracts and invite policy heat. The same tools fueling growth are also drawing scrutiny for drones, missiles, surveillance and hacking, which can change the risk profile faster than the hype cycle.
Uncertain headlines remind us to keep a plan and diversify with patience. Our CEO Jaspreet Singh is hosting a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, on September 29th. Sign up free to join him live.
