Archive · 11 Sep 2026
Anthropic Reports AI-Assisted Spying by China and Iran
Anthropic makes an AI system called Claude.
The company said some governments and government-linked groups used Claude to help spy on people.
The reported targets included democracy supporters, ethnic minorities and people who oppose governments.
Anthropic said these activities came from China, Iran and west Africa.
In one case, Iranian users used AI to find people through their social-media accounts.
In another case, a contractor connected to Mali’s national security authorities used Claude to build intelligence software.
Anthropic also accused two Chinese AI companies of secretly sending many customer questions to Claude to improve their own systems.
The company said it stopped these activities and also blocked some attempts involving weapons, biological research and scams.
Anthropic said the purpose of some biological-research activity was not clear and did not identify the scientists involved.
Anthropic said it disrupted state-sponsored surveillance operations using its AI models between January and July.
The reported campaigns originated in China, Iran and west Africa, targeting dissidents and ethnic minorities.
Iranian actors used AI to identify people through social-media accounts, while a Malian contractor developed intelligence-gathering software.
Anthropic accused Moonshot and DeepSeek of secretly using Claude to generate answers and distill data into their models.
The company also blocked activity involving weapons design, biological research and dating scams, though intent was unclear in some cases.
- Who
- Anthropic reported activity involving state-sponsored or state-aligned actors from China, Iran and west Africa, as well as Chinese developers associated with Moonshot and DeepSeek.
- What
- The company said its Claude AI models were used for surveillance, intelligence software, model distillation, weapons-related activity, biological research and scams.
- When
- The incidents were identified and disrupted between January and July; Anthropic made the disclosures on a Thursday.
- Where
- The reported activity originated in China, Iran and west Africa and targeted communities across Asia and Iranian opponents abroad; the report came from Anthropic in San Francisco.
- Why
- According to Anthropic, the actors used AI to support spying, intelligence gathering, model development and other potentially harmful activities.
Anthropic’s allegations
Reported actors’ positions
Use of AI for surveillance
Anthropic’s allegations
Anthropic said state-sponsored or state-aligned actors used Claude to target dissidents and ethnic minorities, build intelligence tools and identify people through social media.
Reported actors’ positions
The article does not provide responses from China, Iran, Mali or the accused actors, so their positions are not established.
Use of Claude for model distillation
Anthropic’s allegations
Anthropic accused Moonshot and DeepSeek of routing user requests through fraudulent accounts to generate answers and improve their own AI models, potentially exposing sensitive information.
Reported actors’ positions
The article does not include statements from Moonshot or DeepSeek confirming or denying the allegations, nor does it establish whether customers were notified.
Biological-research activity
Anthropic’s allegations
Anthropic said it detected and blocked research involving chikungunya, highly pathogenic bird flu, smallpox- and mpox-related virus families, venoms and toxins.
Reported actors’ positions
Anthropic said the intent was unclear, that the people involved were working scientists, and that it did not assert they intended harm.
- Company
- Anthropic, the San Francisco-based AI lab
- AI model
- Claude
- Reported period
- January through July
- Reported origins
- China, Iran and west Africa
- Reported targets
- Pro-democracy figures in Hong Kong, Tibetan and Falun Gong communities, and Iranian minority communities and regime opponents abroad
- Moonshot activity
- Nearly 300,000 customer requests were reportedly relayed to Anthropic over ten days through 5,380 fraudulent accounts
- Other blocked activity
- Weapons design, questionable biological research and dating scams