September 12, 2026 – AI providers, governments and civil-society organizations should treat automated political profiling and transnational-repression support as an urgent safety problem.
Anthropic reports that it disrupted accounts used for surveillance operations targeting dissidents, ethnic and religious communities, political figures in the Republic of China in Taiwan, labor and student activists, foreign media and human-rights organizations. The company’s attribution confidence varies by cluster, so those assessments should not be overstated.
Immediate actions
- AI providers should strengthen behavior-based detection for bulk political profiling, covert targeting and evasion.
- Providers should preserve relevant evidence and share indicators through lawful, privacy-protective channels.
- Rights groups and journalists should receive clear routes for reporting suspected AI-enabled impersonation or surveillance.
- Public disclosures should protect victims while distinguishing confirmed facts from attribution assessments.