Anthropic Reports Blocking AI Attempts to Enhance Viral Threats

Anthropic revealed that it has successfully blocked multiple attempts by malicious actors to leverage its artificial intelligence models for harmful purposes. The incidents included efforts to facilitate cyberattacks, conduct surveillance, and perform genetic research that could potentially result in the creation of biological weapons. This disclosure comes as part of the company's third report on misuse since March 2025, aiming to shed light on the most novel and significant threats identified to date.

The company emphasized that as AI models become more powerful, the barrier to entry for creating sophisticated threats has lowered. Individuals without advanced technical skills can now generate risks that were previously impossible for lone actors to achieve. In response, Anthropic has implemented stronger safeguards in its latest models to restrict access to biological research that has dual-use potential, balancing scientific advancement with safety.

Intercepting dangerous biological research

One specific incident involved a user attempting to use the Claude model to assist in writing a grant application for scientific funding. The proposed research focused on gain-of-function studies for the chikungunya virus, a mosquito-borne pathogen known for causing severe pain and fever. The goal of the research was to enhance the virus's transmissibility and its ability to evade the immune system.

While such genetic modifications could theoretically lead to better vaccines or treatments, Anthropic noted that the same techniques could be used to make the pathogen more dangerous. The company blocked the request, citing the potential for the research to be weaponized. This case illustrates the dual-use nature of modern biological research, where the line between legitimate science and harmful application is increasingly blurred by AI assistance.

Safeguards face rising capabilities

Anthropic stated that none of the reported misuse cases involved its newest, most powerful model classes, with one exception regarding illicit distillation. The older models, such as Claude Opus 4 and Claude Sonnet 4.5, were considered too limited to meaningfully assist sophisticated users in dangerous biological work. Consequently, their safeguards were primarily designed to prevent novices from accessing known bioweapon information.

However, the company admitted that current models are capable of assisting in complex scientific tasks, and it can no longer guarantee they will not be misused for harm. This shift requires a more rigorous approach to safety. The report urges governments and other AI developers to identify and prevent similar abuse, recognizing that the risks associated with increasingly capable models will continue to grow without collective action.

Industry tensions and transparency

The publication of this report occurred shortly after a researcher resigned from the company, citing concerns that Anthropic and its competitors are not acting responsibly in AI development. This resignation reflects broader industry tensions regarding the pace of innovation versus the speed at which safety measures are implemented. The researcher echoed fears about the technology's potential to elude human control.

According to GN technics/ai (en-US), Anthropic maintains that it has a responsibility to disclose malicious misuse of its services. The company argues that transparency is essential for society's defenders to keep up with evolving threats. By sharing details of these incidents, including snippets of malicious code and prompts, Anthropic aims to foster a more secure environment for AI development and usage.