Anthropic has reported new cases in which its artificial intelligence models were targeted for cyberattacks, surveillance, propaganda campaigns and potentially dangerous biological research, highlighting growing concerns about the misuse of increasingly capable AI systems.
The company said on Thursday that it had blocked attempts by malicious actors to use its models for harmful activities. Anthropic said advances in AI mean sophisticated cyberattacks can increasingly be carried out with less technical expertise, allowing individuals and small groups to create threats that would have been difficult to execute only a year earlier.
The findings were published in Anthropic’s third report on AI misuse since March 2025. The company said the cases were among the most significant and unusual examples of malicious activity it had identified.
The report included examples of malicious code and prompts detected by Anthropic. The company urged governments and other AI developers to monitor their own systems for similar patterns.
Among the cases was an attempt to use Claude to assist with a grant application involving gain-of-function research on chikungunya virus. The research sought to examine mutations affecting the virus’s transmissibility and ability to evade the immune system.
Anthropic said such research can have legitimate medical purposes, including the development of vaccines and treatments, but could also potentially be used to increase the danger posed by a pathogen.
The company said its older models were considered less capable of meaningfully assisting sophisticated users with dangerous biological research. However, it said the capabilities of newer systems have changed the assessment and made stronger safeguards necessary.
Anthropic has therefore introduced tighter restrictions covering a wider range of biological research queries involving dual-use applications in its newer models.
The company said the incidents identified between December 2025 and August 2026 involved actors including spyware vendors, politically motivated individuals and state-sponsored groups. It also identified nine operations involving hundreds of fake social media accounts created to appear as ordinary users while promoting coordinated political messages.
The campaigns originated in countries including Russia, Iran and Turkey, as well as parts of the Gulf, South Asia, Africa and Europe.
Anthropic said AI systems could provide an early warning because the activity may become visible while an influence campaign is still being developed rather than after material has spread widely online.
The report was released two days after Anthropic researcher Jacob Coxon announced his resignation over concerns about the direction of AI development. Coxon accused Anthropic and rival OpenAI of moving rapidly toward self-improving AI systems while failing to adequately address potential risks.
Anthropic said all malicious activities described in its report had been blocked. It said information from the cases had been shared with government authorities and industry partners and used to improve its safeguards.
The company said increasingly capable AI systems require stronger protections from developers, governments and wider society to prevent emerging threats from causing harm.
