A Bombshell Threat Report
In a detailed threat intelligence report released in September 2026, the AI safety and research company Anthropic disclosed that it had detected and disrupted multiple attempts to misuse its powerful AI model, Claude. The report, which covered activities
from late 2025 to August 2026, was a sobering look at how advanced AI is being leveraged for potentially malicious purposes. While the documented cases spanned everything from cyberattacks and surveillance to state-sponsored propaganda, the most alarming section detailed what Anthropic called “biological misuse”—efforts that could directly support the development of bioweapons.
The Five Alarming Cases
The report outlined five distinct instances where working scientists used Claude for research with potential bioweapon applications. In one particularly concerning case, a scientist used the AI to help write a grant application for research on the chikungunya virus. Anthropic noted this was especially worrying because the research was connected to a military research institute. Other cases involved studying highly pathogenic avian influenza and its adaptation to mammals, research into orthopoxviruses (the family that includes smallpox), and compiling data on paralytic venom toxins under the guise of developing new painkillers. After detecting the activities, Anthropic banned the user accounts involved.
The Dual-Use Dilemma
What makes this issue so complex is the dual-use nature of the research. The very same information that could help create a deadly pathogen could also be used to design a life-saving vaccine. Anthropic stressed that its systems cannot reliably determine a user's true intent based on technical queries alone, making it difficult to distinguish between legitimate science and malicious work. The company chose not to identify the scientists and acknowledged that it could not definitively prove they intended to cause harm. However, the nature of their queries and, in some cases, their attempts to bypass safeguards by accessing Claude from unsupported regions, triggered internal alarms and led to intervention.
A Constantly Escalating Challenge
The report highlights a critical shift in the AI landscape. Anthropic stated that while its older models were not powerful enough to meaningfully assist a sophisticated user in dangerous biological research, it can no longer make that same promise for its newer, more capable models. This increasing power has forced the company to implement much stronger safeguards and restrict a wider range of queries related to dual-use biology. The incidents also revealed how determined users try to get around safety measures, for example by routing a request that was blocked by one model to another, more permissive AI. This cat-and-mouse game underscores the reality that AI companies are now on the front lines of preventing the misuse of their own powerful creations.
More Than Just a Biology Problem
The threat report made it clear that the risk extends far beyond bioweapons. Anthropic also detailed how state-linked groups from China, Russia, and Iran used Claude for other alarming activities. These included developing software for conventional weapons like guided rockets and autonomous drones, creating surveillance systems to monitor populations, and running sophisticated influence campaigns. From simulating attacks on Taiwanese air defences to profiling dissidents, the report paints a picture of AI as a powerful tool that is actively being integrated into global conflict and espionage. Each case provides a stark example of how AI's helpful capabilities can be turned toward harmful ends.
