FOLLOW US

Key Takeaways

• Anthropic's newest threat intelligence report details five case studies in which researchers allegedly tried to enlist Claude models in biological weapons-adjacent work, including gain-of-function studies on chikungunya and bird flu.
• The company says internal safety classifiers caught the suspicious activity, but it acknowledges that separating legitimate vaccine research from weapons development is often a judgment call rather than an obvious red flag.
• Anthropic banned the accounts involved but is withholding the names of the scientists and institutions, saying it has no proof of malicious intent and doesn't want to expose working researchers to unwarranted fallout.

By Samuel López | USA Herald

Anthropic has revealed that it identified and shut down several scientists who allegedly tried to use its AI models to advance biological weapons research, according to a threat intelligence report the company published Thursday. Biological misuse is only one category the report covers, but it's drawing outsized attention because Anthropic's newer models, including Fable 5, have grown noticeably more capable at assisting with serious scientific work — the same capability that makes them useful for both a vaccine breakthrough and something far more dangerous.

The report walks through five case studies involving Anthropic's models and biological weapons development, along with the internal safeguards that caught the activity and how the company responded. Anthropic told The New York Times that catching this kind of misuse is genuinely difficult, since research aimed at stopping a virus can look nearly identical, on paper, to research aimed at weaponizing one. In each instance, the company says it chose to act cautiously given how high the stakes could be.

"It's an incredibly nuanced situation," Jacob Klein, Anthropic's head of threat intelligence, told The Times, explaining that the company isn't dealing with cartoonish villains announcing an intent to cause mass harm, but rather requests that require careful, contextual judgment calls before anyone sounds an alarm.

One case from May stands out. Anthropic's biological safety classifier flagged a request asking Claude to help draft a grant proposal for gain-of-function research on chikungunya, a virus with no licensed treatment that can leave patients debilitated for weeks or months. The proposal reportedly sought to make the virus both more transmissible and better at evading the immune system — and Anthropic says its ties to a military research institute made the request even more concerning.

A similar pattern emerged around gain-of-function research involving bird flu. In a separate case, the report says a researcher used Claude to help build an atlas of venom toxin peptides alongside a generative pipeline designed to optimize toxin characteristics — the kind of dual-use project that could serve legitimate research or something far more troubling, depending on intent. Beyond biological research, the broader report also documents Anthropic models being misused for surveillance, building software exploits, generating propaganda, and assisting with weapons systems design.

Anthropic says every account tied to the misuse has been banned, and the findings are now feeding into stronger safeguards meant to catch similar behavior earlier. Notably, the company is not naming any of the individuals or institutions connected to the biological weapons case studies. "The individuals implicated in these case studies are working scientists," Anthropic said in the report. "We do not assert that they intended harm, and identifying them or their labs could expose them to harm."

That distinction matters. Anthropic isn't describing a foiled terrorist plot — it's describing working scientists whose research requests tripped safety systems built to catch a much darker possibility. For a company racing to build increasingly powerful research tools while also policing how those tools get used, the report is as much an admission of how blurry that line can be as it is a demonstration of the guardrails meant to police it.

Stay Ahead of the Story

Want investigative reporting like this delivered before anyone else has it? USA Herald's newsletter brings you exclusive legal analysis, national security coverage, and insider-level reporting you won't find anywhere else. Subscribe to the USA Herald newsletter and never miss a story.

👉 Follow us on X: https://x.com/RealUSAHerald

About the Author

Samuel López is an independent news reporter for USA Herald, where he has covered legal and national security affairs since 2022. A seasoned legal professional with more than two decades of experience across the legal and insurance sectors — most recently as a Senior Legal Analyst — López brings an investigative journalist's eye and a legal researcher's rigor to complex litigation and policy stories.