The Anthropic AI misuse report details how criminals, state-sponsored groups, spyware vendors, scientists, and propagandists tried to weaponize its Claude models. The company published the findings Thursday.
The 154-page document lays out threats across biology, cyberattacks, surveillance, and propaganda. Anthropic said it banned the accounts involved. The company called it the most detailed account of Claude misuse it has released.
The Anthropic AI misuse report frames these cases as rare but serious. The company said it acted to protect the public.
“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,” Anthropic wrote. “We’re publishing this work because we believe we have a responsibility to disclose malicious misuse of our services.”
Biological research raises the loudest alarm
The Anthropic AI misuse report highlights five biological research cases. The company’s safety teams flagged each one.
Anthropic said some researchers dodged safeguards meant to block users from “unsupported regions.” Others hid the purpose of their work.
“Biological misuse is one of the most serious risks of frontier AI models,” Anthropic wrote. “Without the correct safeguards, such capabilities could have catastrophic consequences.”
The Anthropic AI misuse report describes one scientist who used Claude on a state-sponsored grant application. The work studied chikungunya, a mosquito-borne virus that causes severe pain, fever, and other debilitating symptoms. The virus resembles dengue and malaria.
Such research could help build vaccines. It could also help build weapons. Anthropic said the study’s tie to a military research institute deepened its concern. Anthropic did not name the institutions or countries, citing uncertainty about intent.
Claude surfaces in weapons and surveillance cases
The Anthropic AI misuse report also documents conventional weapons work. Threat actors used Claude to develop software for firearms, missiles, armed drones, bombs, and other munitions.
Anthropic traced this activity to Yemen, China, and Russia. The cases spanned weapons design, guidance systems, and procurement.
The Anthropic AI misuse report points to Russian espionage and “smash-and-grab” cyberattacks. In some operations, AI helped attackers automate reconnaissance, exploitation, and data theft. The tools let small crews do work that once needed larger teams.
Surveillance drew fresh alarm. Anthropic reported a China-based program that used Claude to target Uyghurs in Syria. The company said the effort aimed to track, profile, and recruit its targets. Another operation tracked internal dissidents. Propaganda campaigns surfaced in Russia, Malaysia, Iran, and Bangladesh.
Report lands amid warnings from inside Anthropic
The Anthropic AI misuse report arrived two days after a company researcher resigned. Jacob Coxon said he quit because Anthropic was not acting responsibly. His resignation set off a media firestorm.
Coxon said Anthropic and its rival OpenAI were “racing straight to self-improving superintelligence.” He warned the technology could cause human extinction by 2030. Some current employees publicly agreed.
The Anthropic AI misuse report offers a different lens on AI danger. It focuses on present harm, not a future machine that escapes control.
Many experts say near-term threats matter more right now. Heidy Khlaaf, chief AI scientist at the AI Now Institute, pressed that point.
“AI accelerationism and AI doomerism are two sides of the same coin: they both enforce the notion that an AGI superbeing will come into existence,” Khlaaf said.
She said labs that build tools for cyberattacks and warfare could prove far deadlier than any distant superintelligence.
Anthropic calls for shared defenses
The Anthropic AI misuse report says most real-world misuse stays out of public view. Companies, academics, governments, and international organizations investigate it privately.
Anthropic urged the industry and governments to work together. It said defenses must keep pace as models grow stronger.
“As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer,” Anthropic wrote.
The Anthropic AI misuse report makes one point clear. Misuse is no longer a theoretical policy debate. It is an operational test of the safeguards around powerful AI.
What worries you more: near-term misuse of AI tools or the long-term risk of systems slipping beyond human control? Please share your view in the comments below.