Anthropic report details blocked attempts to use Claude AI for bioweapons, kamikaze drones, and Russia-linked espionage against Ukraine

Anthropic's 154-page threat report reveals it blocked users in Houthi-held Yemen and others from using Claude to develop biological weapons and guided munitions.

Anthropic has published a 154-page threat intelligence report documenting repeated attempts by malicious actors to exploit its Claude AI models for dangerous purposes — including efforts that could have contributed to the development of biological weapons. The report, released on a Thursday, represents one of the most detailed public disclosures by a major AI company of the specific ways bad actors are attempting to weaponize commercial AI systems. Among the cases highlighted, users located in Houthi-controlled territory in Yemen attempted to use Claude to develop guided weapons systems. The Houthis, who control large parts of Yemen including the capital Sanaa, have been engaged in an active armed conflict and have conducted drone and missile attacks on regional shipping lanes and neighboring countries. Additionally, Anthropic said it disrupted a suspected Russia-linked cyber espionage campaign that used Claude to target Ukrainian government, military, and diplomatic sectors. Russian developers were also found to have used the AI to help build software for 'kamikaze' attack drones. Beyond these cases, Anthropic said it blocked a range of other misuse attempts, including broader cyberattacks, surveillance operations, and research that could have supported the creation of biological weapons. The company categorized those responsible as 'bad actors' operating both domestically and abroad. The report arrived two days after a notable internal development: former Anthropic employee Jacob Coxon publicly resigned, citing concerns that the company was accelerating development of AI models he believed could contribute to human extinction by 2030. Anthropic has not publicly addressed Coxon's specific claims in connection with the report. The disclosures underscore a growing tension in the AI industry between the rapid commercial deployment of powerful language models and the difficulty of preventing their misuse by state-aligned groups, criminals, and other bad actors who can access them through standard interfaces.

Why it matters

The report demonstrates that commercial AI systems are already being tested as tools for weapons development by active armed groups, raising urgent questions about how AI companies can enforce safety guardrails at scale. The breadth of misuse attempts — spanning bioweapons research, cyberattacks, and guided munitions — signals that AI safety is no longer a purely hypothetical concern.

What's next

The full 154-page report is publicly available at Anthropic's website, and policymakers and AI safety researchers are likely to scrutinize its findings as Congress and regulators debate AI oversight frameworks.

Key facts

Bias & framing notes

Most outlets reported the story factually and consistently, but framing varied: The Washington Post's headline emphasized Houthi use as a confirmed fact ('Houthis used Anthropic'), while PBS and the New York Times used more cautious language ('blocked possible efforts'). The Guardian uniquely highlighted the Coxon resignation as context, which other outlets omitted. The Western Journal emphasized the 'bad actors' and bioweapons angle with a more alarming framing. No outlet provided Anthropic's stated explanation for why it chose to publish the report now.

NewsClear — neutral news & congressional tracking · Bill of the Week