Anthropic says Claude was used in missile guidance and espionage operations

Anthropic has said its Claude artificial intelligence models were used in attempts to support missile guidance software in northern Yemen and to run a Russian-linked cyber-espionage operation. The company set out the allegations in a new threat report published on Friday, describing what it said were multiple malicious campaigns that it disrupted. It said the cases ranged from weapons design to phishing, data theft and covert influence activity.

Sponsored

In the Yemen case, Anthropic said it intervened to block an effort to deploy Claude for missile guidance software, including work linked to a guided rocket and a long-range ballistic missile. The report said the operators used the model in place of human software engineers and assigned different instances of the system separate roles for missile-guidance and flight-control code. Anthropic said many requests were stopped by its safeguards, but that some prompts still got through.

It also said the operators tried to avoid detection by hiding their end goal and splitting tasks across multiple sessions. The company said it has no evidence that the group succeeded in fielding a working weapon. It added, however, that the operators appeared to have carried out an unsuccessful test-fire.

Anthropic said it banned the accounts involved and shared threat information with public- and private-sector partners to reduce the risk of similar activity. The report also said the company disrupted a Chinese operation and removed three Iranian state-aligned accounts, although those cases were separate from the Yemen allegation. The report is significant because it points to the growing use of generative AI in both military and intelligence-related activity.

Sponsored

If confirmed, the Yemen case would suggest that AI tools are being tested not only for ordinary software work but also for weapons development. The espionage allegation is also notable because Anthropic said the Russian-linked operation relied on automated AI workflows for much of its activity, including phishing, setup and data theft. Anthropic said the Russian-linked campaign bore the hallmarks of Midnight Blizzard, also known as APT29, and targeted Ukrainian, European and diplomatic interests, including drone makers.

The company said the operation used AI to handle nearly the entire process, which it described as a shift from isolated assistance to more automated offensive activity. It also said the Chinese operation used Claude as an engineering and orchestration layer for attacks on government and corporate networks across the Middle East, Europe and Southeast Asia. What remains unclear is how far any of the reported operations progressed before being stopped, and whether any weapons-related work in northern Yemen moved beyond testing.

Anthropic said it had blocked many requests, but acknowledged that some had slipped through. The company said it had banned the relevant accounts and continued monitoring for similar behaviour, while the wider question now is how quickly such systems can be adapted for offensive use and how effectively they can be contained.

#Anthropic #Claude #missileguidance #cyberespionage #Yemen

Sponsored