Is AI Being Misused for Bioweapons, Espionage and Mass Surveillance?

Artificial intelligence is increasingly being used for sophisticated cyber and intelligence operations, raising concerns about the misuse of powerful AI models. Anthropic, the company behind Claude, has revealed several cases in which its AI systems were allegedly misused for espionage, surveillance, cyberattacks, weapons-related activities and potentially dangerous biological research.

According to Anthropic's latest threat intelligence report, malicious actors and state-backed groups used Claude models for a range of activities. The company said it identified and disrupted several such operations over the past eight months.

AI Misused for Biological Research

One of the major concerns highlighted in the report is the potential use of AI in biological research that could have dangerous consequences. Anthropic said it identified attempts to use its models for research involving viruses and gain-of-function studies.

In one case, a user sought help preparing a grant application related to gain-of-function research involving the chikungunya virus. Such research can have legitimate scientific purposes but can also raise serious safety concerns when it involves increasing a pathogen's ability to spread or evade immunity.

Anthropic said it blocked accounts linked to suspicious activities and has strengthened safeguards in its newer AI models.

AI Used in Espionage and Cyberattacks

The report also highlights the growing use of AI in cyber operations. According to Anthropic, attackers are moving beyond using AI simply as a tool for individual tasks. AI agents are increasingly being used to coordinate multiple stages of cyberattacks, including reconnaissance, data theft and other activities with limited human involvement.

Anthropic also reported a suspected Russia-linked cyber espionage campaign in which AI was used to support operations targeting organisations in Ukraine.

Surveillance Operations

Another major concern is the use of AI for mass surveillance. Anthropic said government-linked actors in countries including China, Iran and Mali used Claude to assist with activities such as collecting and analysing information and identifying potential surveillance targets.

The company said it took action against accounts connected to such operations and introduced additional security measures.

AI and Weapons Development

Anthropic also identified cases involving the use of Claude in weapons-related activities. The report said actors in countries including China, Russia and Yemen attempted to use AI for weapons development and related intelligence and procurement work.

The findings have added to concerns that increasingly capable AI systems could lower the barriers to accessing expertise in sensitive areas such as cyberwarfare, weapons development and biological research.

Anthropic said the incidents show why AI companies need stronger safeguards and why governments and technology firms must work together to prevent the misuse of advanced AI systems.