Anthropic Reveals AI Misuse in Cybercrime and Surveillance

Anthropic has detailed a series of malicious activities involving its Claude AI model, ranging from industrialised cyberattacks to automated surveillance and influence operations.

In a recent report, the AI safety and research company identified various ways in which its technology has been misused. The findings suggest a pattern where human actors select targets and set strategic goals, while AI agents perform the high-volume, technical tasks required for reconnaissance, exploitation, and data theft.

Automation of Cybercrime and Phishing

The report describes how attackers are leveraging AI to scale traditional cybercrime methods. This includes the automation of credential theft, cloud compromise, and phishing campaigns. AI is also being used to accelerate vulnerability research and the extraction of sensitive data from downstream organisations.

These developments follow previous concerns regarding Anthropic AI models being used to target organisations and broader safety risks inherent in large language models.

Influence Operations and Surveillance

Beyond direct cyberattacks, the company documented the use of AI in influence operations. These activities involved the creation of synthetic personas, fabricated journalists, and fake news sites to produce large-scale, multilingual content. While these operations generated high volumes of content, the report noted that they often achieved limited genuine engagement.

In the realm of surveillance and repression, the misuse included the production of automated dossiers and the analysis of biometric and communications data. The report also highlighted instances of transnational targeting and coercive recruitment. Notably, some surveillance systems were found to continue operating locally even after the model’s access had been revoked.

Dual-Use Biological Risks

The report also addressed the dual-use risks associated with biological and military applications. While AI has been used to support advanced scientific and military research, Anthropic stated that it has not established evidence of completed biological weapons or the operational deployment of AI in battlefield settings.