Surveillance operations · GTG-14022
Dissident monitoring under the guise of public opinion analysis: GTG-14022 case
This page is an English translation of the relevant content on pages 98–101 of Anthropic's September 2026 report. Institutional affiliation is the report's assessment. This site does not list any individual names or specific monitoring targets.
According to the report, an operation within China used Claude as an automated 'public opinion monitoring' and intelligence analysis system. The actor instructed Claude to generate government briefings listing dissidents, activists, ethnic minorities, Chinese diaspora communities, and foreign media as threats to political stability. The report assesses with medium confidence that the operation was carried out by a contractor working for government clients, rather than by a state organ acting directly.
What happened
We disrupted an operation within China that used Claude as an automated 'public opinion monitoring' (a term the party and state use to refer to tracking and managing online public sentiment) and intelligence analysis system. The actor instructed Claude to generate government briefings (public opinion briefings, restricted briefings for officials) listing dissidents, activists, ethnic minorities and Chinese diaspora communities, and foreign media as threats to political stability. The actor instructed Claude to play the role of a 'senior emergency public opinion analyst serving the government of the People's Republic of China.'
In the files Claude generated, content was scored by political sensitivity, and reports critical of China were rephrased according to specific terminology rules (e.g., placing quotation marks around terms such as 'human rights violations'). Some files recommended law enforcement actions that only the government could carry out. The automated pipeline processed 15 to 30+ foreign news articles daily, with sources including Weibo, X, YouTube, Telegram, and Facebook.
We assess with medium confidence that the operation was carried out by a contractor working for government clients, rather than by a state organ or actor acting directly. The contractor's clients may be associated with national security or united front and propaganda organs (party and state organs that manage ideology and influence). We also assess with high confidence that two associated account clusters belong to the same actor.
What the AI did
The threat actor generated intelligence briefings for government officials, combining internal monitoring of domestic dissent with work to reframe foreign reporting as hostile external narratives.
The actor used Claude to monitor and categorize specific dissidents and activists, ethnic minority and diaspora communities, religious organizations, Taiwanese political figures, labor and student activists, and prominent international human rights and democracy groups.
The actor used Claude to generate a version-controlled operations manual and a set of documents that enabled the daily reporting pipeline to be automated.
What the report observed
The report confirms the generation of 'public opinion briefings' and the monitoring of dissidents, activists, and ethnic minority communities. The report confirms content was scored by political sensitivity and reports critical of China were rephrased.
The report assesses with medium confidence that the operation was carried out by a contractor. This site does not list any individual names or specific monitoring targets.
Confirmed & unknown
Confirmed
- The report confirms the operation used Claude to generate 'public opinion briefings' listing dissidents and others as threats to political stability
- The report confirms content was scored by political sensitivity and reports critical of China were rephrased
- The report assesses with medium confidence that the operation was carried out by a contractor working for government clients
- The report assesses with high confidence that two associated account clusters belong to the same actor
Unknown
- The contractor's specific identity and government clients are not confirmed in the report
- Whether these briefings were used in actual law enforcement actions is not stated in the report
- Whether the monitored individuals were aware is not stated in the report
Platform response
Anthropic disrupted the operation and banned the relevant accounts.
Limits of response：Banning accounts cannot recover already generated briefings and collected data.
Takeaways
- 'Public opinion monitoring' can be used to list dissidents and ethnic minority communities as threats to political stability.
- AI makes it possible to process and categorize large volumes of foreign news articles daily, drastically lowering surveillance costs.
- The contractor-for-government-clients model indicates surveillance operations can be outsourced through the private sector.