AI company Anthropic has revealed how governments, criminal groups and other actors allegedly misused its Claude AI models for activities ranging from biological research and weapons development to mass surveillance, espionage, cyberattacks and influence operations.In a safety report covering incidents detected between December 2025 and August 2026, Anthropic said the cases involved suspected state-backed groups, financially motivated criminals, propaganda networks, spyware vendors and politically driven actors.The company said Claude Haiku, Sonnet and Opus were used in the cases outlined. None involved Claude Fable or Mythos-class models, except for one distillation case. Anthropic said the examples represented some of the most notable and novel threat activity it had identified.Claude and biological researchAnthropic said some of its most sensitive cases involved biological research by working scientists, including users it described as potentially state-supported. The company said it could not always determine whether the work was legitimate or intended for harm.Jacob Klein, Anthropic’s head of threat intelligence, said the distinction was not always straightforward.“You are not seeing someone in a comic book kind of way say, ‘Hey, I want to build a biological weapon to kill everybody,’” Klein said. “It’s an incredibly nuanced situation.”Anthropic said some users circumvented controls restricting access from countries where Claude is unavailable and attempted to conceal the purpose of their research. The company said it banned the accounts and, in biological cases, used hard refusals, weaker model tiers and proactive monitoring.AI-assisted weapons developmentAnthropic also identified attempts to use Claude to develop software for conventional weapons, including firearms, missiles, armed drones and bombs. The company said it detected three cases in China, two in Russia and one in Yemen.In Yemen, Anthropic said a group used Claude Code in place of human guidance, navigation and control engineers while developing software for a guided rocket, a multistage ballistic missile intended to exceed 2,000 km in range and a hypersonic glide vehicle variant. The company said the group test-fired the guided rocket, but the field test appeared to have failed.In Russia, Anthropic said a freelance operator linked to an effort called DronDoc or Serafim used Claude Code to engineer an autonomous first-person-view kamikaze drone swarm. According to the company, the system could select targets, including a “person” class, and detonate without a human in the loop.In China, Anthropic said an account potentially linked to the military-industrial sector used Claude to build a 16-module electronic warfare and air-defence suppression suite before shifting the simulation to 12 real targets in Taiwan.Surveillance and espionageAnthropic recorded nine cases involving alleged mass surveillance. In one, a group suspected of links to China allegedly used Claude to track, profile and recruit Uyghur populations and journalists with ties to the Syrian Army.The company also described China-based surveillance activity targeting Catholic cardinals, the Presbyterian Church in Taiwan, Tibetan Buddhists and Falun Gong.In Iran, Anthropic said two linked units using 16 Claude accounts claimed to have surveilled or profiled 6,388 Iranians over a year. They allegedly analysed 155,216 tweets to identify 39 opposition accounts and used a malicious Firefox extension to collect identities in a shared case-management system called Arman.On cyber operations, Anthropic said AI had helped actors automate reconnaissance, exploitation and monitoring. One Russian-speaking actor allegedly targeted more than 20 Ukrainian and European government, defence and diplomatic organisations and drone manufacturers.Anthropic said the actor stole a drone vision-system software development kit, hijacked hotel Wi-Fi DNS records to plant malware and obtained more than 300,000 national identity records and over 500,000 company registry entries from a North African government body.Influence operationsAnthropic said it disrupted at least nine influence operations involving Russia, China, Iran, Bangladesh and Kenya. The groups allegedly used Claude to plan campaigns and generate false or misleading content.Among the cases were Russian state-media insiders, including a former Sputnik Moldova editor-in-chief, who allegedly used Claude as a sub-editor. Anthropic also said a Russian-speaking coordinator in Bangui used Claude for Radio Lengo Songo, a Wagner-founded station, to generate pro-Russia and anti-France content and forge Central African Republic government documents.Data breaches and fake dating profilesAnthropic said financially motivated groups also misused Claude. In one case linked to ShinyHunters affiliates, operators allegedly downloaded 1.8 million Android application packages and scanned them for hardcoded secrets.The company also detected a China-based network of more than 20 dating applications marketed as “fully human” but largely powered by Claude-driven personas. Over two weeks, Anthropic identified more than 4,700 AI personas, which sent 2.36 million messages to at least 25,000 real users.Rival AI companies and ‘illicit distillation’Anthropic also accused rival AI companies and related networks of using Claude outputs to train their own models, a practice it calls “illicit distillation”.It named campaigns linked to Alibaba, Moonshot AI, DeepSeek, Zhipu, Xiaomi, SenseTime and MiniMax.The company said the largest measured campaign, attributed to Alibaba, reached nearly three million exchanges a day at its peak and more than 151 million exchanges between May and July 2026 through more than 3,500 fraudulent accounts.Anthropic said it responded by banning accounts, tracing proxy networks, strengthening extraction detection and requiring identity verification for accounts showing abuse signals.AI safety warnings intensify in USThe report comes as concerns about AI safety are also growing in the US.Anthropic researcher Jacob Coxon recently resigned, saying that “people building AI earnestly believe that it could kill us all by the end of the decade.”Anthropic scientist Evan Hubinger echoed the warning, writing on X: “We really do earnestly believe AI could kill all humans. I personally think it is >10% within the next decade.”The warnings have prompted calls from US lawmakers for stronger safeguards. Democratic Senator Mark Kelly said, “Washington needs to wake up and take this seriously,” while Republican Senator Ted Cruz has pointed to legislation addressing AI’s “catastrophic risks”.California has also enacted a law setting rules for independent audits of AI products. Meanwhile, lawmakers are working on proposals that would require developers of the most powerful AI models to undergo independent security audits.Anthropic said it was publishing its findings because misuse would grow as AI models become more capable unless developers and defenders work to make them safer. The company said the report could help other platforms recognise similar patterns and strengthen collective defences.Inputs from Reuters