An AI Company Says It Stopped AI From Helping Make Bioweapons
Anthropic says it stopped bad actors from using its AI in ways that could have helped make biological weapons.
The company described the cases in its September 2026 threat-intelligence report, a public list of harmful uses it blocked. The report covers December 2025 to August 2026 and lists seven kinds of harm, from hacking to biological misuse. Reporting on Sept. 11 said the blocked biological work could have supported weapons development, news that drew international attention PBS.
After the report, experts called for stricter rules on who can use powerful AI, pointing to more misuse cases tied to biology research Al Jazeera. They did not ask for a small fix. They asked for limits on what AI can do, how it is shared, and who can use it.
This follows earlier warnings. Scientists gave The New York Times transcripts where AI chatbots described how to build deadly germs and release them in public The New York Times. In another case, germ and terrorism experts who read ChatGPT chats called some answers deadly accurate The Wall Street Journal.
A new study raised more concern that AI could help make a new kind of biological weapon The New York Times. That connects to a debate in Nature about whether to limit AI used for biology, since AI can design viruses, poisons and other bioweapons. Researchers also found that AI-designed poisons can slip past safety checks used by companies that sell genes, much like a fake ID slipping past a door check.
Governments have watched this risk for years. In July 2023, Democratic and Republican U.S. senators both warned about AI, focusing on the risk of a biological attack. U.S. intelligence later gave its formal view. The ODNI 2026 Annual Threat Assessment states that AI has the potential to aid in weapons and systems design.
In June 2026, Anthropic shared a proposed plan for governments to handle four risk areas for AI, including biological weapons. In an hourlong interview, Bill Gates said separately the tech industry was downplaying AI risks, which he said include mass job loss and bioterrorism The New York Times.
The broader context here is a change from what might happen to what was tried and stopped. Early warnings asked what AI could do. Now reports list what people tried, what was blocked, and where safety checks failed. That moves the question to how to respond, share information, and set lines for action.
Looking at what this means for countries working together, splits are likely. Nations will disagree on how to check compliance, how to control AI business and gene sales, and how much warning data companies should share. Companies see the use. Governments make the rules. Scientists know the germs. No one can fix this alone. The test is whether they can agree on the facts before the next report shows a more advanced try.
