OpenAI and Anthropic urge US DNA screening law to curb AI bioweapon risk

This is one of the clearest examples yet of frontier AI labs trying to move the policy battlefield from abstract “AI risk” debates to a very specific chokepoint in the bio supply chain. By backing mandatory screening and record‑keeping for synthetic DNA and RNA, Anthropic, OpenAI, Microsoft AI and others are effectively saying: we can’t reliably control what our models teach people, so we should harden the physical layer instead. It’s notable that key gene‑synthesis vendors are co‑signers—this isn’t just AI companies trying to regulate someone else’s industry.

For the race to AGI, the letter is less about slowing capabilities and more about keeping catastrophic misuse from derailing progress. If AI makes it easier for non‑experts to design dangerous pathogens, the political backlash could quickly translate into far broader bans on frontier models or compute. Narrow, enforceable safeguards at the DNA ordering layer are a comparatively surgical intervention that lets labs keep scaling while addressing the most acute tail risks.

Competitively, labs that actively shape workable biosecurity regimes position themselves as responsible partners to governments, which can pay dividends when access to sensitive data, regulated sectors, and national compute resources are on the table. Those that sit it out may find future rules written without them—potentially in ways that favor their more proactive rivals.