Anthropic, the U.S. firm behind the Claude artificial‑intelligence chatbot, disclosed that its technology has been exploited by a range of hostile actors to develop missile guidance software, conduct cyber‑espionage and even pursue biological‑weapon research. The company said it identified and blocked more than a dozen coordinated misuse campaigns spanning Yemen, Iran, China, Russia and other jurisdictions.
Missile Guidance and Battlefield Applications
According to reports from Al Jazeera, The Times of India and India Today, a Houthi‑affiliated weapons cell in Yemen used Claude to design and test long‑range missile guidance algorithms. The same AI tool was allegedly employed by Iran‑linked accounts to track U.S. naval vessels, a claim echoed by Iran International. These findings suggest that advanced language models can lower the technical barrier for creating sophisticated weaponry, a concern highlighted by multiple outlets.
State‑Backed Espionage and Propaganda
Anthropic’s internal September 2026 report, referenced by Reuters and the Korea Times, said Chinese and Russian state actors attempted to harness Claude for intelligence‑gathering, targeting software and disinformation campaigns. Chinese AI labs, including Alibaba and Moonshot AI, were accused of siphoning millions of Claude exchanges to train their own models without permission, a claim also reported by CNBC. The company said it disrupted these illicit activities after detecting anomalous usage patterns.
Unauthorized Model Training by Foreign Labs
Chinese‑based entities allegedly used Claude’s output to accelerate their own generative‑AI development, a practice described by CNBC as “unauthorized efforts.” Anthropic said it detected large‑scale data extraction attempts and took steps to revoke access for the implicated accounts. Similar concerns were raised by Devdiscourse, which noted that U.S. lawmakers are urging a blacklist of certain Indian IT firms over alleged hacking activities linked to AI misuse, though Anthropic did not confirm any direct involvement of Indian companies.
Blocking Bioweapon Development
The BBC and CNN reported that Anthropic blocked attempts to employ Claude for the design of biological weapons. In a statement quoted by the BBC, the company said:
"We have blocked malicious use of our AI that could develop biological weapons."Anthropic added that a nine‑month sweep uncovered nine separate inquiries from actors seeking assistance with bioweapon research, all of which were denied.
Anthropic said its detection systems flagged the suspicious activity through a combination of usage‑pattern analysis, user‑behavior monitoring and collaboration with external security partners. The firm has since tightened its API controls, introduced stricter identity‑verification procedures and expanded its “red‑team” testing to anticipate novel abuse scenarios, as detailed in its public “Detecting and Countering Misuse of AI” brief.
Industry analysts note that the disclosures come as Anthropic prepares for a high‑profile initial public offering. Whalesbook observed that the company’s willingness to publicize the misuse cases may be aimed at reassuring investors and regulators about its governance framework. Nevertheless, the revelations have intensified calls for tighter AI regulation, with several European and U.S. lawmakers citing the incidents as evidence that existing safeguards are insufficient.
Anthropic’s experience underscores a broader trend: generative‑AI tools, once limited to text generation, are now being repurposed for military, intelligence and illicit scientific applications. As governments grapple with the dual‑use nature of these technologies, the firm’s latest report is likely to shape forthcoming policy debates on export controls, licensing and international cooperation to curb AI‑enabled threats.
