Anthropic Says It Shut Down Several Efforts to Use Claude to Develop Biological Weapons

Anthropic disclosed in its "Detecting and countering misuse of AI" threat intelligence report, published Thursday, that it detected and shut down state-sponsored efforts to use its Claude model to research biological weapons, including possible mutations of the chikungunya virus, between December 2025 and August 2026. The company said this is the first known public disclosure by a company of AI models being potentially misused for biological weapons, and that in one case the same actors re-established new accounts within days after being intercepted. Anthropic also reported disrupting state-aligned influence operations and cyberattacks, saying "sophistication has stopped being a reliable signal of who is behind an operation.

In a new report discussing safety measures that have been taken by Anthropic when it detects that its models are being used for potentially malicious activity, the company says its AI was used to possibly begin developing new bio-weapons. It could be part of a concerted PR push to counter the narrative, which was amplified in a former employee's public resignation letter https://sfist.com/2026/09/09/anthropic-researcher-says-theres-10-chance-of-ai-killing-all-humans-within-10-years/ this week, that AIs will probably kill us all before long. But Anthropic is effectively saying, "Hey We've got this handled," via its latest, albeit very unsettling report on " Detecting and countering misuse of AI https://www.anthropic.com/threat-intelligence-report-september-2026?ref=sfist.com ," published Thursday morning. Though the report may raise more concerns and questions than it does alleviate them. The report covers potential threats that were detected and disrupted between December 2025 and August 2026 — the company's previous, similar threat report came out in November 2025, and highlighted how they had disrupted the first reported AI-orchestrated cyber espionage campaign https://www.anthropic.com/news/disrupting-AI-espionage?ref=sfist.com by a state-sponsored group in China. The latest report details some similar types of cyberattacks, with the topline there being that AI has enabled far less sophisticated, non-state-sponsored bad actors to conduct the kinds of attacks that, just a year ago, would have required far more well developed hacking know-how. "For threat intelligence investigators, sophistication has stopped being a reliable signal of who is behind an operation," the report's authors say. "Every layer of offensive operations has been uplifted by AI, from reconnaissance and tool development to data processing and exploitation." The company also disrupted some state-aligned political influence and fake-news operations, the likes of which we've seen for a decade but which are now super-charged using AI-generated personas, biographies for fake reporters, impersonations of real-life figures, and more. But buried a hundred or so pages into this lengthy report, Anthropic explains that, since last November's previous report, the company has detected new categories of threats related to state-sponsored use of its AI for the creation and control of conventional weapons, drones, weapons for undersea warfare, and biological weapons. Some of these efforts are linked to specific countries, including China and Russia. But when the report gets into the section on biological weapons, the authors demur and decline to identify the countries that specific state-sponsored actors came from whose activity they shut down. Anthropic says that this is the first time that they are aware of in which a company has publicly disclosed detection of the potential misuse of their AI models for biological weapons. In one instance, the company says that it found a state-sponsored military effort, coming from a region where Claude would normally be blocked from use, to research possible mutations of the chikungunya virus, a mosquito-borne virus that causes debilitating symptoms that can last weeks or months, and that, because it comes from insects, a deliberate deployment might be indistinguishable from a natural outbreak. Chillingly, after the first activity was intercepted, the same actors re-established new accounts and continued the research within days, trying to evade detection. And it sounds like Anthropic continues to engage in a Whac-a-Mole effort as new accounts and methods arise from this same "cluster of activity." In May, Anthropic said it disrupted some research being done by an actor within the US into highly-pathogenic avian influenza or bird flu, specifically seeking genetic variants with pandemic potential. Avian influenza variants currently show very low ability to pass from human to human, but a variant with this capability could be massively deadly. The New York Times spoke today https://www.nytimes.com/2026/09/10/us/politics/anthropic-ai-biological-weapons.html?ref=sfist.com with microbiologist Dr. Susan Monarez, who briefly served as CDC director last year, about the Anthropic report. Adding another check in the "it's going to kill us all" column, Monarez expressed her concern that advanced AI models may "let bad actors hide in plain sight, using seemingly legitimate research to create pathogens we may not see coming and may not be able to stop once released." Anthropic notes in its report that it takes seriously how it examines biological research activity, because the same research that could be used for good, say in the development of new vaccines, could also be used for malicious purposes. Related: Anthropic Researcher Says There’s 10% Chance of AI Killing ‘All Humans’ Within 10 Years https://sfist.com/2026/09/09/anthropic-researcher-says-theres-10-chance-of-ai-killing-all-humans-within-10-years/