Anthropic intervened in multiple attempts by scientists and other actors who were using its models to help develop biological weapons, the artificial intelligence research company said in a report released Sept. 10.
The lengthy document details how Anthropic found and prevented various cases of its AI models being used maliciously between December 2025 and August 2026.
“The cases we share here aren’t typical misuse, but rather examples of the most notable and novel threat activity we’ve identified to date,” it said. “As models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer.”
In addition to biological weapons, Anthropic noted six other areas of harm it disrupted, including cyber operations, influence operations, surveillance, scams and fraud, conventional weapons development and illicit distillation, which is the unauthorized replication of one model’s capabilities into another model.
Anthropic did not respond immediately to a request for comment from USA TODAY.
CANADA – 2026/08/27: In this photo illustration, the Anthropic AI logo is seen displayed on a smartphone screen. (Photo Illustration by Thomas Fuller/SOPA Images/LightRocket via Getty Images)
The report cited biological misuse as “one of the most serious risks of frontier AI models.” In the biological misuse cases, researchers could not confirm whether the work was legitimate or with bad intent, so it shut down suspicious users who evaded safeguards and, in some cases, appeared linked to military research.
Doni Bloomfield, a Fordham law professor who focuses on AI and biosecurity, told USA TODAY that he found the use cases linked to military complexes most interesting, though “it’s not clear how much the models that were accessed at their best, could have even done. But I think it is notable that this work is happening.”
Five case studies were presented in the paper to demonstrate how Anthropic AI could be used in developing bioweapons. One plays out a scenario in which a user asked the model to write an application for a grant to research orthopoxvirus at a state-associated infectious disease lab. Orthopoxviruses are associated with smallpox and Mpox, which caused a global outbreak in 2022.
The goal of the research was to better understand genes that are immune to the virus, which could be used by someone to preserve, enhance or transfer virus immunity.
“You also have to pair this with advances that we’re seeing in other forms of AI,” Bloomfield said. “It’s a number of these advances happening simultaneously that are especially concerning. And it’s not especially reassuring that these are some of the requests that they’re getting.”
Based on its findings that today’s models might be able to “meaningfully” help carry out dangerous biological research, Anthropic relaunched its most recent modes with stronger safeguards, according to the report.
“This is an indication that as these models get a lot better, we need to be quite thoughtful about how we regulate the safeguards that are in place,” Bloomfield said.
Greta Reich covers the artificial intelligence industry for USA TODAY through a fellowship from the Tarbell Center for AI Journalism. Funders do not provide editorial input.
This article originally appeared on USA TODAY: Anthropic reports it blocked potential bioweapons research
Leave a Reply