Anthropic released a report on September 10, 2026, describing how it detected and blocked misuse of its Claude AI models. The report covers activity between December 2025 and August 2026.
The company said it disrupted cyberattacks, surveillance operations, and attempts at biological research that could support weapons. Anthropic shared details with government authorities and industry partners in some of the cases.
This is the third report of its kind from Anthropic since March 2025. The company said it hopes other AI developers and governments can use the findings to spot similar patterns.
Cyberattacks Grow Easier to Launch
Anthropic said AI has narrowed the gap between skilled state-backed hackers and less experienced individuals. Tasks that once required a team of specialists can now be handled by one person with AI help.
One case involved a group Anthropic linked to Russian state-backed hacking activity known as Midnight Blizzard. The group used Claude to automate phishing, malware development, and data theft against Ukrainian government and military targets.
The same group also went after European diplomatic missions and companies tied to drone supply chains. Anthropic said the group used AI to rebuild its tools automatically whenever security software detected them.
Another case involved hackers connected to the ShinyHunters group, known for large data breaches followed by extortion demands. Anthropic said these hackers used Claude to scan millions of app files for stolen passwords and access tokens.
The hackers targeted a technology provider, an airline, and an energy company, among others. In some cases, the stolen data included passenger records and customer payment information.
A separate case involved operators based in China who used Claude to research and build exploits against security software. Anthropic said two of the people involved were university students in China's Hunan province.
Concerns Over Biological Research
Anthropic said it also blocked a request tied to research on the chikungunya virus, a mosquito-borne illness that causes fever and joint pain. The request sought help writing a grant application for research aimed at making the virus more transmissible and better able to evade the immune system.
Anthropic said this type of research could help develop vaccines and treatments. It also said the same research could make the virus more dangerous if misused.
The company said its newer models, including Claude Fable 5, now carry stronger limits on biological research requests. Anthropic said its older models from 2025 were not capable enough to meaningfully assist with dangerous biological research.
Anthropic also said it found nine cases of coordinated social media accounts spreading political messaging. These accounts originated in countries including Russia, Iran, and Turkey.
The report was published two days after Anthropic researcher Jacob Coxon announced his resignation. Coxon said he was concerned that Anthropic and other AI companies are moving too fast toward more powerful AI systems.
Anthropic said it banned the accounts linked to the misuse cases described in the report. The company said it used what it learned to strengthen safeguards across its models.
Anthropic is preparing for an initial public offering later this year. The company said it plans to keep publishing updates on AI misuse as part of its ongoing safety work.
