In a report published on Thursday, September 10, Anthropic disclosed information about 39 cases of Claude being used for malicious and criminal purposes, recorded between December 2025 and August 2026. These range from fraud, identity theft, or API key theft, to espionage attempts and the development of military or biological weapons1.
- Claude has notably been used by Iranian, Russian, and Emirati state actors to carry out propaganda campaigns, build surveillance tools targeting minorities and dissidents, and disseminate content favorable to their regimes.
- For example, a Russian state actor used the chatbot to produce pro-Russian and anti-French propaganda content broadcast via a Central African Republic radio station, and to fabricate fake Central African administrative documents.
- The company also revealed that Emirati officials used Claude to orchestrate a campaign against the Muslim Brotherhood by relying on fake influencers and on the impersonated identity of a Swiss NGO, which was used to publish fake human rights reports authored by Abu Dhabi.
Accounts linked to the Iranian regime have likewise attempted to deploy Claude to mount influence campaigns aimed at foreign and Iranian public opinion.
- One of these actors notably used the chatbot to produce official documents as part of these campaigns, and devised an organizational plan for the funeral of Supreme Leader Ali Khamenei.
- Another actor asked Claude to reproduce the voices of Iranian writers and to prepare in advance speeches for the succession of the Supreme Leader.
In several cases documented by Anthropic, Claude was used to conduct surveillance and espionage operations, or to develop guidance software for rockets and missiles. The company says it “ceased the activity in every instance, strengthened our AI protections based on lessons learned, and shared intelligence with authorities and industry partners when appropriate.”
- Chinese actors used Claude in attempts to monitor, profile, and recruit Uyghur-origin individuals who had joined the Syrian Army, offering money in exchange for information about their units.
- In Russia, actors used Claude to develop a swarm of autonomous FPV kamikaze drones. The vision system was trained on footage taken on the front in Ukraine.
- Finally, actors based in northern Yemen, a region partially controlled by the Houthis, attempted to use AI to develop a guided rocket, a ballistic missile with a range over 2,000 kilometers, and a family of missiles including a hypersonic glide vehicle variant.
Anthropic also acknowledged the use of its chatbot by several researchers in attempts to develop biological weapons.
- The company cataloged five cases in which Claude was used for these purposes: a grant application for research on the chikungunya virus intended for a military institute, work on adapting avian influenza H5 to mammals, a grant application on immune evasion by orthopoxviruses, and two government-backed programs for the design of venoms and toxins.