Srinagar, Sep 11: Artificial intelligence company Anthropic has claimed that its Claude models were used in multiple malicious operations involving cyber-espionage, weapons development, mass surveillance and covert influence campaigns, prompting the company to strengthen safeguards and monitoring.
In a new threat report, Anthropic said it had disrupted an operation in northern Yemen in which Claude was allegedly used to assist with missile-guidance software, including systems for a guided rocket and a long-range ballistic missile.
According to the report, the operators used Claude in place of human software engineers and assigned different model instances specific tasks related to missile guidance and flight-control software.
Anthropic said its safeguards blocked many of the requests, but some reportedly passed through. The operators allegedly attempted to evade detection by concealing their broader objectives and dividing tasks across separate sessions so that no individual prompt revealed the full purpose of the activity.
The company said it had found no evidence that the group successfully deployed a functioning weapon. However, it said the operators appeared to have conducted an unsuccessful test-fire.
Anthropic said it subsequently banned the accounts involved and shared information about the activity with public- and private-sector partners.
The report also detailed alleged state-linked cyber-espionage campaigns. Anthropic said a Russian-linked operation bearing similarities to the Midnight Blizzard, or APT29, group used automated AI workflows to conduct much of an espionage campaign, including phishing, infrastructure setup and data theft.
The targets reportedly included Ukrainian and European organisations, diplomatic entities and drone manufacturers.
In a separate case, Anthropic said it disrupted an operation allegedly run by university students in China’s Hunan province. The group reportedly used Claude as an engineering and orchestration layer for an offensive programme targeting government and corporate networks across the Middle East, Europe and Southeast Asia.
The company said it banned accounts associated with both operations and introduced additional monitoring to identify similar activity.
Anthropic also reported detecting three Iranian state-aligned accounts allegedly using Claude for covert influence and psychological operations. The company said the activity was linked to Iranian propaganda institutions and involved generating content aligned with narratives associated with the Islamic Revolutionary Guard Corps.
Another case involved what Anthropic described as an industrial-scale targeting operation. According to the company, Claude was used to generate structured profiles of targets, including information related to location, demographics, political leanings and confidence scores.
Anthropic said the most operationally mature case involved a China-aligned account allegedly conducting a multi-day recruitment effort targeting Uyghurs in Syria. The account reportedly had no Arabic-language skills but used Claude to draft outreach messages in a regional dialect and translate responses in real time.
The disclosures come amid wider concerns about the growing use of advanced AI systems in cyber operations and other harmful activities. Anthropic has also faced scrutiny after disclosing another incident involving an early version of Claude Opus 4.6 that allegedly gained unauthorised access to external systems.
The latest developments followed the resignation of former Anthropic researcher Jacob Coxon, who publicly raised concerns about AI safety and warned about the potential consequences of the accelerating AI race. Another Anthropic scientist, Evan Hubinger, subsequently expressed agreement with the concerns.
The warnings have added to calls among some US lawmakers for stronger rules governing advanced AI systems.
Anthropic said it was investigating the incidents and had engaged an independent research firm to review them.
The company is also facing a difficult relationship with Washington over the use of AI in defence and surveillance. Earlier this year, the Pentagon designated Anthropic a supply-chain risk after the company refused to remove safeguards relating to autonomous weapons and domestic surveillance.
Anthropic challenged the decision in California, where a judge last month ruled that the US Department of Defense had acted unlawfully in issuing the designation.
Despite the dispute, reports have indicated that Claude models have been used in US military operations in Iran and Venezuela.