Bill Gates warns AI has crossed risk thresholds in biology, cyber and jobs

Bill Gates published today the first document in a planned series in which he argues that current frontier models have already passed points of no return in five risk categories: biological capabilities, cyber capabilities, psycho-social capabilities, destruction of the labor market and loss of control. In an interview with MIT Technology Review the former Microsoft founder said he is “simply stunned by the lack of concern and discussion outside the industry” and described himself as “a harsh voice” trying to wake decision-makers.

Biological capabilities at the top of the list

Gates’ primary concern is the ability of models to generate new molecules. “Any model that can generate new molecules must be under surveillance,” he said, estimating that bioterror risk is fifty times higher than the risk of a natural pandemic. He added that guardrails intended to prevent misuse are not keeping pace with model progress, and the threshold at which access could be stopped or limited has already been breached.

Policy proposals: reserved jobs for humans and tax on robots and tokens

To address the economic implications, Gates advances two ideas he has raised before. The first, “reserved jobs for humans,” would require companies to keep certain roles filled by human employees, with the exact list determined separately in each country. The second, a tax on robots and tokens, would channel part of the revenue generated when AI replaces human labor to the public treasury, echoing an automation tax he proposed years ago.

Cautious optimism: agriculture, health, education and bureaucracy

Despite the warnings, Gates remains optimistic about long-term benefits. He points to precision agriculture, medical diagnosis, personalized education and navigation of government bureaucracy as areas where the technology already creates tangible value. “In the end we will reach abundance,” he concluded, “but first there will be upheaval. And a lot of it.”

Why now

Gates says two factors drive the timing: the crossing of the technical thresholds described above, and the realization that earlier promises – “when we get close to the threshold we will know how to allow only good use” – have not materialized. The current document is the first in a series, and the public and private interviews he is conducting with industry, government and civil-society leaders are intended to push for regulation before the gap between capability and control widens further.