Anthropic Says It Blocked Hackers and Researchers From Misusing Claude for Bioweapons, Cyberattacks and Spying

Anthropic says it caught and stopped a series of attempts to turn its Claude AI models into tools for bioweapons research, cyberattacks and spying. The company laid out the cases in its latest threat intelligence report, published this week, which covers suspicious activity detected between December 2025 and August 2026. The report paints a picture of how fast criminals, state-linked groups and researchers are testing the limits of frontier AI, and of how much work goes into keeping those models out of the wrong hands.

The most serious cases involve biology. Anthropic says it found five cases in which scientists used Claude in ways that could potentially support the development of biological weapons. In one case, a researcher got around geographic restrictions and used Claude over several weeks to plan experiments involving the adaptation of avian influenza to mammals. Anthropic says it eventually spotted the activity and banned the accounts involved. The company stressed that the cases do not show Claude successfully creating a biological weapon, but the attempts themselves were enough to trigger a ban.

What the report says about cyberattacks

Anthropic also says it found suspected state-linked hackers using Claude to support cyber operations. One group showed techniques consistent with a Russia-linked threat actor previously associated by US authorities with Russian intelligence. According to Anthropic, the group wove AI into multiple stages of cyber operations aimed at Ukrainian government, military and diplomatic organizations.

The reported uses covered the full playbook of a modern hacking campaign. The attackers reportedly used Claude for phishing, for developing malware, and for rewriting malicious code when security software caught it. Anthropic also described hotel Wi-Fi hijacking and WhatsApp account takeover operations in the same campaign, with AI used at nearly every stage. Reuters reported that Anthropic disrupted the related activity, and the Russian Embassy did not immediately respond to a request for comment.

The report goes further than one campaign. Anthropic says it also found attempts to use Claude for conventional weapons development, including software related to missiles, armed drones and targeting systems. Activity connected to actors operating in Russia, China and Yemen appeared among the cases the company investigated.

The Alibaba distillation case

A separate part of the report deals with model theft, known as distillation. This is the practice of using the outputs of a powerful AI system to help train another model without permission, effectively copying its abilities on the cheap. Anthropic says it accused operators linked to several China-based AI laboratories of trying to extract knowledge from Claude this way.

The largest case described in the report was attributed to Alibaba. Anthropic says it observed more than 151 million exchanges between May and July 2026, at times approaching three million per day, across more than 3,500 accounts it considered fraudulent. The company alleged that the activity aimed to improve Alibaba's Qwen models. Reuters said Alibaba did not immediately answer a request for comment on the claims.

Anthropic says it has also been fighting distillation for a while. In a separate announcement in October, the company said new safeguards had cut the amount of reasoning data harvested in distillation attacks by about 90 percent to 99 percent, and that it had banned tens of thousands of accounts linked to such campaigns. The new report shows the problem has not gone away, even if the defences have improved.

Influence operations and fake accounts

The report also covers political influence operations. Anthropic says it discovered operations creating hundreds of social media accounts designed to look like ordinary users and distribute coordinated political messaging. The accounts were built to blend in, which is exactly what makes them hard for platforms and readers to spot.

This is part of a wider pattern the company describes. Threat actors now use AI across large portions of an operation, not just for one task. Models can help write code, analyse information, organise campaigns and adapt when something fails. That makes a small team far more productive, and it makes the early stages of an attack, which used to take real skill and time, much cheaper to run.

What Anthropic did about it

Anthropic says it banned every account connected to the activities it identified. It also strengthened its safeguards, built new detection methods and shared relevant information with government authorities and industry partners, according to the Associated Press. The company says most of the biological cases involved older systems, including Claude Opus 4 and Claude Sonnet 4.5.

The report is also a reminder that these findings come from Anthropic itself. The attributions, including the link to a Russia-linked group and the case against Alibaba, are the company's own claims, not independent court judgments. That does not make them unimportant. Anthropic is one of the few companies that can see this activity at scale, because it runs the systems being misused, and its reports are one of the main public windows into how AI misuse actually works.

For ordinary users, the report carries a practical message too. The same AI tools that help with writing, coding and research can be turned toward phishing emails, scam messages and fake accounts. If a message looks slightly off, or a stranger pushes you to act fast, slow down and check it through another channel. The technology behind the scam is getting better, and the companies building it are now telling us openly that the fight to keep it in check is already under way.

Frequently Asked Questions

What is in Anthropic's latest threat report?

The report covers malicious activity detected between December 2025 and August 2026, including five cases linked to possible bioweapons research, state-linked cyber operations, conventional weapons work, political influence operations and large scale attempts to steal model capabilities through distillation.

Did Claude actually help build a bioweapon?

No. Anthropic stressed that the cases do not demonstrate that Claude successfully created a biological weapon. The company found scientists using Claude in ways that could potentially support bioweapons development, including planning experiments on avian influenza, and it banned the accounts involved.

What did Anthropic do with the accounts it caught?

Anthropic says it banned every account connected to the activities it identified, strengthened its safeguards, created new detection methods and shared relevant information with government authorities and industry partners.