Reuters, citing Anthropic’s threat intelligence report, stated that Claude has been exploited by various organizations and individuals in activities related to weapons development, cyberattacks, surveillance, and fraud before the accounts were detected and handled.
Anthropic accused certain actors in China of using Claude to develop electronic warfare software, simulate radar jamming, and plan air defense suppression, including the simulation of 12 targets in Taiwan such as early warning radars, Patriot and Sky Bow missile sites, air bases, and command bunkers.
Another group was allegedly using Claude to build a technical proposal of over 200 pages for the Chinese Navy’s anti-torpedo system and compare it with US Navy technology.
The report also described Claude being used to research high-power microwave weapons, trace component supply chains, and prepare reports for Chinese agencies.
In Yemen, Anthropic discovered a group using Claude to program, simulate, and troubleshoot guided missiles, ballistic missiles with a range of over 2,000 km, and a variant utilizing a hypersonic glide vehicle. The company stated that many requests were blocked and there is no evidence these weapons have been deployed in practice.
Certain actors in Russia were accused of using Claude to develop control software for autonomous FPV drone swarms and attempting to procure dual-use components through intermediaries in China and Hong Kong to evade sanctions.
Anthropic noted numerous efforts to use Claude to support biological research related to the chikungunya virus, highly pathogenic avian influenza, orthopoxviruses, and novel toxins, while stating that these activities took place in regions where its services are not provided.
In the cybersecurity field, the company stated that Claude was used to find software vulnerabilities, develop exploit code, and assist in infiltrating the networks of approximately 50 organizations, including foreign government agencies.
The report also detailed Claude being used in intelligence-gathering operations, surveillance of Uyghurs, Tibetans, activists, Jewish organizations, and US naval forces, as well as building social media monitoring platforms and a tracking system for about 25 million SIM cards in Mali.
Anthropic also discovered a campaign in China developing over 20 dating apps featuring more than 4,700 interactive AI personas targeting at least 25,000 users to facilitate fraud operations.
📌 Anthropic’s report shows that generative AI is increasingly being exploited for various security and military purposes, ranging from weapons development and biological research to cyberattacks, surveillance, and fraud. These are allegations and assessments made by Anthropic based on detection activities within its systems; some of the accused governments and organizations have denied the claims or have not yet responded. The content emphasizes the need to strengthen oversight on the use of advanced AI models to mitigate the risk of misuse.
