In a STAT opinion piece, Ashish K. Jha, former White House Covid-19 response coordinator, argues that AI is eroding the barriers that historically kept biological weapons rare. He points to Anthropic's recent report of five cases where its models were used for work that could support bioweapons development, including proposals for gain-of-function research and experiments to help bird flu infect mammals. Anthropic blocked the accounts but could not determine intent.
Jha says AI models trained on biological literature can connect ideas across subfields, helping someone with training in one area learn key steps in another. He cites two reports, one from Harvard and one from RAND, that independently concluded AI could broaden the range of actors able to mount a large-scale biological attack and make existing state programs more capable.
The second barrier is DNA synthesis. Jha cites a Nature Communications paper in which researchers ordered genetic fragments of the 1918 influenza virus and 36 of 38 DNA synthesis companies sent the sequences. Because shorter fragments remain unregulated, a skilled person could reconstruct the virus. Jha argues that even perfect screening would not be enough, since DNA synthesis is a global market.
Jha calls for a layered defense, including verified access to leading-edge biological models and oversight, but warns that broad restrictions would slow biomedical progress. He concludes that the threat landscape is moving faster than defenses.