Anthropic reports Chinese state actors used Claude to surveil dissidents and religious groups
Anthropic has disclosed that Chinese state-linked intelligence actors abused its AI assistant, Claude, to conduct large-scale surveillance on dissident communities and religious groups. The company’s September 10 threat intelligence report details how these actors used the model to compile dossiers on targets including Falun Gong practitioners, Tibetan Buddhists, and employees of NTD, a sister…
Key points
- Anthropic banned Chinese state-linked accounts that used Claude to compile dossiers on dissidents and religious leaders.
- Actors used the AI to process multilingual data from social platforms and generate daily Chinese-language intelligence briefings.
- The surveillance targeted groups including Falun Gong, Tibetan Buddhists, and NTD employees, replacing traditional analyst teams.
The operation involved a user identifying as a Chinese state information security officer who managed a “religious affairs intelligence collection desk.” This actor utilized Claude to run four simultaneous work streams, processing multilingual data from platforms like WeChat, Douyin, and LinkedIn to generate Chinese-language briefings. The AI was instructed to extract sensitive personal details, such as immigration dates, and map religious locations using structural diagrams. Anthropic noted that this activity replaced the need for traditional analyst teams, highlighting the efficiency with which state actors can now leverage generative AI for espionage.
In response, Anthropic banned the involved accounts and enhanced its detection systems to identify similar patterns. The company warned that such sophisticated misuse of AI safeguards is likely to increase as the technology advances. This incident underscores the growing geopolitical tensions surrounding AI capabilities, where powerful models are being repurposed for state-sponsored repression and intelligence gathering against specific political and religious minorities.
Chinese state-linked actors used AI assistant Claude to target dissidents, religious groups
bing.com · 14 September 2026Beijing [China], September 14 (ANI): Chinese state-linked actors used Anthropic’s artificial intelligence assistant Claude to monitor dissident and religious communities, including Falun Gong practitioners and employees of The Epoch Times (TET) sister outlet NTD, according to tech company Anthropic, as reported by TET.According to Anthropic’s Sept. 10 threat intelligence report, the actors targeted communities that are suppressed by Beijing and used the AI model to compile dossiers on religious leaders and identify possible vulnerabilities or pressure points that the Chinese regime could exploit. Anthropic said the Chinese actors were using AI to carry out tasks that would traditionally require a team of analysts. The company described the activity as part of efforts by sophisticated threat actors to test AI safeguards, warning that such risks are likely to grow as the technology continues to advance, as highlighted by TET.The report said that one individual operating an account identified himself as an information security officer working for the Chinese state. Anthropic assessed the accounts as being connected to a China-based intelligence operation, noting that their activities closely followed the priorities of China’s religious affairs apparatus, the country’s top intelligence agency, and the United Front Work Department, the party-state organization responsible for co-opting the Chinese diaspora and targeting critics of the regime.Anthropic said it banned the accounts involved and enhanced its systems to identify similar activities. In one campaign, the company identified an operator managing what it described as a “religious affairs intelligence collection desk.” The user employed Claude to launch four simultaneous work streams, where the AI model processed information in several languages, translated the material, and generated Chinese-language dossiers, reports, and daily briefings, TET reported.The operation focused on senior Catholic cardinals across Asia, leadership of the Presbyterian Church in Taiwan, Tibetan Buddhists and leaders living in exile, Falun Gong practitioners, and NTD. The actors gathered information from Chinese platforms including WeChat, RedNote, Douyin, the Chinese version of TikTok, and Weibo, along with Western platforms such as LinkedIn, Instagram, Threads, X, and Facebook.The information collected included sensitive personal details such as birthdays and immigration dates. The actors also attempted to identify and map religious locations using floor plans, images of building facades, and structural diagrams. According to the report, the information-gathering cycle was conducted daily and appeared to be designed for internal state security offices.Anthropic said the actors instructed Claude to gather information about a target’s activities related to China, any “negative information,” as well as personal or organisational weaknesses that could potentially be used as leverage, as cited by TET.Falun Gong, also known as Falun Dafa, is a spiritual practice that focuses on meditative exercises and the principles of truthfulness, compassion, and tolerance. It was first introduced publicly in China in 1992 and rapidly expanded through word of mouth, reaching an estimated 70 million to 100 million practitioners by the end of the decade.The Chinese Communist Party launched a nationwide campaign of persecution against Falun Gong in July 1999. Since then, millions of practitioners have faced arbitrary detention, forced labour, and torture, while some have also died as a result of forced organ harvesting, according to TET. (ANI)(This content is sourced from a syndicated feed and is published as received. The Tribune assumes no responsibility or liability for its accuracy, completeness, or content.)
This text was published by bing.com and written by ANI. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.
More in Policy & Regulation
All →- OpenAI pledges to match Anthropic's embedded evaluator safety commitment · 92 src
- Meta Faces Class Action Over AI Training and Facial-Recognition Claims · 1 src
- Sanders proposes 20-year prison term for developing Artificial Superintelligence · 5 src
- FTC Ex‑Commissioner Dismisses AI Doomsday Claims, Urges Coordination · 1 src
- AI Snake Oil argues for AI control over alignment to prevent agent loss-of-control · 1 src
Comments
via GitHub Discussions