Anthropic says it blocked AI misuse for weapons and espionage

Anthropic says it has disrupted attempts to use its Claude artificial intelligence models for activities that could support biological weapons development, missile software, cyber-espionage, propaganda, surveillance and fraud.

The California-based company detailed the cases in a threat intelligence report released Thursday, saying it believes it has “a responsibility to disclose malicious misuse of our services.” The activity it described spanned suspected state-sponsored groups, criminals, spyware vendors, state propaganda institutions and politically motivated individuals.

According to the BBC, Anthropic said it disrupted “malicious use” of its Claude Haiku, Sonnet and Opus models between December 2025 and August 2026. None of the cases involved Claude Fable or its more powerful Mythos-class models, except for one instance of distillation, a process used to train smaller AI models from larger ones.

Anthropic said it blocked scientists who used Claude in ways that could support biological weapons development, highlighting “five case studies of actors using our models in ways that could support biological weapons development.” The company called biological misuse “one of the most serious risks of frontier AI model,” warning that without proper safeguards such capabilities “could have catastrophic consequences.”

“The same information that can be used to develop a biological weapon could also be used to develop, for example, a vaccine or a cure for a disease,” Anthropic said. Jacob Klein, the company’s head of threat intelligence, told The New York Times it was “an incredibly nuanced situation.” He added: “You are not seeing someone in a comic book kind of way say, ‘Hey, I want to build a biological weapon to kill everybody.’”

The report also cited six cases in which Claude was used “to develop software for conventional weapons, including firearms, missiles, armed drones, bombs, and other munitions, as well as the targeting and control systems that operate them.” Al Jazeera reported that Anthropic intervened in northern Yemen to block an effort to use Claude for missile guidance software, including for a guided rocket and a long-range ballistic missile. The operators allegedly used Claude “in place of human software engineers,” assigning different model instances to write guidance and flight-control software.

Anthropic said safeguards blocked many requests, but some slipped through because the operators obscured their goals and divided tasks across separate sessions. The company said it had no evidence the group fielded a working weapon, though it claimed the operators appeared to have conducted an unsuccessful test-fire.

The report also described cyber operations. The BBC reported that a hacking group whose work was consistent with Russia-based Midnight Blizzard allegedly used AI to build a system that detected when malware was flagged by security defenses and rewrote code until it evaded detection. Al Jazeera reported that a Russian-linked operation bearing hallmarks of Midnight Blizzard used automated AI workflows for phishing, setup and data theft against Ukrainian, European and diplomatic targets, including drone makers.

Anthropic also said it disrupted a Chinese operation run by university students in Hunan province that used Claude “as the engineering and orchestration layer” for an offensive program targeting government and corporate networks across the Middle East, Europe and Southeast Asia, according to Al Jazeera. The company said it banned associated accounts and added monitoring.

The report said Claude was also used by Iranian state-aligned accounts for covert influence and psychological operations tied to Iranian propaganda institutions, including the Islamic Culture and Communications Organisation and a Mashhad seminary command room distributing content aligned with Islamic Revolutionary Guards Corps narratives. Anthropic also described a China-aligned account with no Arabic skills using Claude in a “multi-day recruitment operation to infiltrate Uyghur targets in Syria,” saying the model drafted outreach in a regional dialect and translated replies in real time.

Anthropic said it has incorporated its findings into its systems to better prevent, detect and disrupt misuse, and has shared intelligence with authorities and industry partners where appropriate.