I wrote about Anthropic’s threat intelligence report two days ago. Since Sunday there is a second half to the story: Science asked virologists and biosecurity researchers what they make of the five biology cases. The answers are far apart.
What the cases were
Three of the five involve viruses. In May, Claude wrote a grant application for a project meant to find mutations in the chikungunya virus that help it spread better and evade the immune system. The prompts made clear the researchers planned to build those viruses and test them in animals. Also in May, Claude helped plan experiments on a highly pathogenic avian influenza strain: mutations that adapt the virus to mammals. The third case involved an orthopoxvirus made less virulent in mice.
Case four was an atlas of venom toxins for a drug development program, built from public data. Case five was the computational redesign of toxins.
Anthropic calls the chikungunya work “highly concerning gain-of-function research” and the avian flu experiments research into a pathogen with enhanced pandemic potential. The latter is no longer permitted in the US under the Trump administration’s new rules.
The pushback
Kristian Andersen, a virologist and evolutionary biologist at Scripps Research, dismissed it on Bluesky: most, if not all, of the experiments are “just basic research.” Talking to Science he gets more specific. These are mutations that already occur in nature. Isolating such viruses alive is hard, since they are usually known only from sequencing, so virologists routinely drop them into a weakened backbone or a pseudovirus that cannot replicate at all. It sounds scary, Andersen says, but it is ordinary biological research and perfectly safe in high-containment labs. He points to a 2023 paper he co-authored: same approach, Zika instead.
Gregory Koblentz of George Mason University makes a similar argument. Every pathogen in the report is endemic somewhere and a current public health threat. Studying them is not suspicious on its own.
Filippa Lentzos of King’s College London refuses both camps: “I would resist both extremes in interpreting these cases.” Gigi Gronvall of Johns Hopkins, who has advised Anthropic but was not involved in the report, calls some of the online reaction “overheated.” Both still think publishing it was useful, because it starts a conversation that had not happened yet.
On the other side is Andrew Weber of the Council on Strategic Risks, who read the report before release. He told the New York Times these are “chilling examples of state-sponsored biological weapons developers.” And Ashish Jha, who helped steer the Biden administration’s COVID-19 response, wrote on X: “We won’t get five more warnings before something bad happens.”
Why this one stuck with me
On Saturday I read the report mainly as a security story. The virologists read it as a technical paper, and there it comes apart.
Which is why Andersen’s real objection is not about virology at all. Companies like Anthropic get to decide for themselves which research they consider dangerous, and then shut it down. That is exactly what happened: accounts banned, access cut. Five research groups whose work at least two experts call routine were pulled off the platform by a vendor in California.
Anthropic had reasons, and good ones: disguised origins, workarounds after the ban, an institute tied to a military lab. That is in the report and it carries weight. But nobody reviewed the decision, and there is no process for reviewing it either. Two days later Dario Amodei named a ban on biological weapons development as something countries might agree on. Until they do, the vendor decides. It is not a good answer. Right now it is the only one on offer.
Sources: