Anthropic Says Bad Actors Tried To Turn Claude Into a Weapon and a Spy

Anthropic PBC has released a threat report stating that bad actors attempted to bend its Claude AI model toward warfighting and espionage. According to the company, users in Iran, Russia, China, and Yemen engaged in activities touching kamikaze drone swarms, missile navigation, and projects connected to potential biological weapons over the past year. Anthropic stated that the incidents did not involve its top-end models, Fable and Mythos.

The report detailed five examples tied to work that could advance biological weapons, including a case where state-sponsored virologists sought an enhanced strain of chikungunya. Anthropic noted that all the biological examples were ambiguous and could reflect legitimate science. On the battlefield side, non-state-affiliated users in Russia reportedly tried to build software to run a swarm of FPV loitering munitions learning from Ukrainian combat footage. In Houthi-controlled northern Yemen, operators turned to Claude to develop programs for several different missile types, with Anthropic stating they did not have evidence the actors succeeded in fielding an operational device, though they did test-fire a guided rocket and returned to Claude to troubleshoot its failure.

Media additions

Separate operations connected to China and Iran aimed to build and enable surveillance. According to the company, a China-linked effort used Claude to track, profile, and recruit Uyghurs in the Syrian Army, while an Iranian unit built surveillance code disguised as a prayer times app. Furthermore, hackers leaned on AI to move faster. Microsoft Corp. Linked a group known as Midnight Blizzard to Russian intelligence services and reported that they increased their speed by automating operations using AI, including using Anthropic's software to automate phases of malware development and craft phishing emails. Anthropic also stated that the ShinyHunters cyber-extortion outfit used AI to scan for leaked credentials to steal data from affected organizations.

These developments coincide with major geopolitical friction involving the United Kingdom, France, Canada, and Israel. On Tuesday, UK Foreign Secretary Ed Miliband announced a comprehensive sanctions regime and a ban on the import of goods produced in illegal Israeli settlements in the occupied West Bank. Miliband stated in the House of Commons that the British government agrees there is ethnic cleansing of Palestinians in areas of the West Bank perpetrated by settler terrorists, and that the occupation is unlawful due to its entrenchment of control and expansionist agenda via illegal settlements. The UK trade ban was announced alongside France and Canada, and nine other European nations signaled support for similar restrictions.

Israel responded swiftly to the UK's announcement. Israeli Foreign Minister Gideon Sa'ar announced the closure of the British consulate in East Jerusalem, declared that British civil servants and military staff would be deported, and that 12 British MPs would be banned from entering Israel. Sa'ar strongly rejected the accusations and argued that the move constituted blatant interference in a sovereign state's electoral process.

Trade data shows that commerce between the UK and Israel totaled 6 billion pounds in 2025, according to House of Commons records cited by Al Jazeera, with an estimated 38 million pounds worth of goods imported from the occupied territory that same year. While the import restrictions represent a small fraction of overall bilateral trade, analysts and officials have debated the enforceability and broader implications of the measures.

The convergence of artificial intelligence risks, military security concerns, and international trade restrictions highlights a complex global landscape. Anthropic's stance on restricting Claude's use for expansive spying or independently operated weapons has strained ties with the Pentagon, which suspended collaboration after the company refused to greenlight such applications.