Ex-official: White House AI accord lacks action triggers

The author's argument rests on a 2023 episode in the White House Situation Room. When H5N1 influenza infected a father and killed his daughter in a remote Cambodian village, the national security adviser asked whether to brief President Biden immediately. The author said to wait, because a trigger had been agreed before the alert arrived: sustained human-to-human transmission. Surveillance data from nearby emergency rooms, compared against prior years, showed cases were not unexpectedly high.

That same standard is what the author says is missing now. Bill Gates has warned that AI crossed thresholds this year, bio-attack capability among them, and described the threshold of most concern as whether an AI can design a bioterrorism weapon capable of killing millions. Anthropic's chief executive urged the U.N. Security Council to back a ban on using AI to make biological weapons. The author notes there is almost no population immunity to H5N1, roughly half of human cases are fatal, and the virus does not yet spread efficiently between people.

The piece draws a distinction between design and physical capability. A model that writes a research protocol to create a pathogen is not a person who can create it; synthesized DNA, equipment and a lab with four walls are still needed to grow a virus. That barrier, not the AI itself, is where the author argues screening and tracing should be expanded, including at DNA synthesis providers, cloud laboratories and benchtop synthesizers.