Anthropic disrupts state-sponsored AI surveillance operations

Anthropic has released a major threat intelligence report detailing the disruption of sophisticated, state-sponsored, and commercial operations misusing its Claude AI models between December 2025 and August 2026. The company blocked a commercial election manipulation platform targeting Malaysian voters linked to Turkey's BBS Bilisim Teknolojileri, state-sponsored surveillance operations from China, Iran, and Mali targeting dissidents, and a Russian-linked cyber espionage campaign against Ukraine. Anthropic also disrupted massive 'illicit distillation' campaigns by Chinese AI labs, including Alibaba and Moonshot, and blocked attempts by scientists to use Claude for high-risk biological research.

Anthropic malicious AI disruption

• ▪Anthropic blocked efforts by bad actors to use its Claude AI models for cyberattacks, conventional weapons development, biological research, and state-sponsored surveillance
• ▪Anthropic published a threat intelligence report on September 10, 2026, detailing its disruption of cyber operations, influence campaigns, surveillance, and other forms of AI misuse between December 2025 and August 2026

Malaysia election manipulation platform

• ▪Anthropic observed a request on the platform for one million artificial views on the account of the sitting Malaysian Prime Minister, but found no evidence linking the operation to Prime Minister Anwar Ibrahim or his office
• ▪The election manipulation platform used Claude to generate fabricated intelligence reports containing false allegations against an unnamed Malaysian opposition politician and civil society organizations
• ▪The Malaysia-focused operation managed over 1,000 fake accounts on X using cookie and IP renewal mechanisms to evade detection, and ran a fabricated news site called Malaysia Pulse to republish rewritten articles
• ▪Anthropic disrupted a commercial election manipulation platform linked to Istanbul-based BBS Bilisim Teknolojileri that used Claude to profile and target voters across all 222 parliamentary constituencies in Malaysia

Chinese AI model extraction

• ▪Chinese AI developers Moonshot and DeepSeek secretly routed live customer conversations through Claude to use Anthropic's responses as training data, exposing sensitive user information to a third party
• ▪Anthropic disrupted an illicit distillation campaign by operators linked to Alibaba that used over 3,500 fraudulent accounts to generate 151 million exchanges between May and July 2026 to improve Alibaba's Qwen models
• ▪Moonshot relayed nearly 300,000 customer requests to Anthropic over a ten-day period using a network of 5,380 fraudulent accounts located primarily in Singapore and Japan

Russia-linked cyber espionage

• ▪Anthropic disrupted a suspected Russia-linked cyber espionage campaign, consistent with the threat actor Midnight Blizzard, that targeted Ukrainian government, military, and diplomatic sectors using AI at nearly every stage
• ▪The suspected Russian hacking group used AI to build a system that automatically detected when its malware was flagged by security defenses and rewrote the code to evade detection

Biological weapons research prevention

• ▪Anthropic blocked five instances of scientists attempting to use Claude for research that could support biological weapons development, including gain-of-function research on the transmissibility of the chikungunya virus
• ▪Anthropic implemented stronger safeguards in its newer models, such as Claude Fable 5, to restrict access to dual-use biological research queries, noting older models were below the threshold of providing meaningful assistance
• ▪The blocked biological research involved working scientists and concerned pathogens such as the chikungunya virus, a highly pathogenic strain of bird flu, and viruses in the smallpox and mpox family

State-sponsored surveillance operations

• ▪In West Africa, a contractor working for Malian national security authorities used Claude to design the underlying software that enabled intelligence gathering
• ▪Anthropic shut down multiple state-sponsored surveillance operations originating in China, Iran, and West Africa that used Claude to target dissidents, ethnic minorities, and pro-democracy figures
• ▪Iranian actors used Claude to develop a method to identify specific individuals through their social media accounts, targeting Iranian minority communities and opponents of the regime abroad

Debatable claims

• ▪Governments should regulate AI safety rather than relying on developer self-policing
• ▪AI developers should restrict access to dual-use biological research queries
• ▪The rapid development of advanced AI models poses an unacceptable risk to human survival