Anthropic says its AI isn't just helping write code and emails—it's now blocking suspected work on biological weapons. In a new report, the maker of the Claude chatbot says it cut off several scientists this year after spotting research requests that could have advanced the design of dangerous pathogens, even though the same techniques can also be used for legitimate vaccine and public health work, the New York Times reports. "It's an incredibly nuanced situation," said Anthropic threat intelligence chief Jacob Klein, who noted users weren't openly declaring plans to build bioweapons.
The report details broader misuse of Claude, including suspected Chinese and Iranian government-linked surveillance of dissidents, Russian state media crafting propaganda disguised as news, and attempts in China, Russia, and Yemen to get help designing conventional weapons software. The most unsettling cases, outside experts say, involve covert biological research, including a grant proposal involving gain-of-function experiments on the chikungunya virus at a military institute. Anthropic reports some users dodged geographic blocks and tried to hide the purpose of their work before being banned.
Biosecurity specialists who reviewed the report argue it's fresh evidence that hostile states and other bad actors are probing advanced AI to boost prohibited weapons programs—and that access to such tools may need to be sharply restricted. Saying it can't be sure of the researchers' intent, the company did not name the institutions or nations where the misuse was attempted, per the Guardian. "Biological misuse is one of the most serious risks of frontier AI models," Anthropic's report says. "Without the correct safeguards, such capabilities could have catastrophic consequences."
,
uBlock 
,
and
Ghostery 
)
