Anthropic CEO Warns of AI Misuse for Bioweapons, Urges Independent Oversight

In a CBS Sunday Morning interview aired on September 13, Anthropic CEO Dario Amodei discussed the risks of AI being misused to develop bioweapons and enhance virus transmissibility. He said he reads threat reports every morning about people attempting to use AI models for such purposes. He noted that Anthropic released a report two days earlier, on September 11, about the misuse of models to build bioweapons.

Amodei argued that a complete ban on AI is not the right path, as it would forfeit enormous benefits including medical breakthroughs. He called for embedding third-party evaluators from nonprofit or government organizations to observe the full process of training and running AI models. He also urged multiple companies to jointly set safety standards, with government involvement in determining release pace. He dismissed the “AI kill switch” as the sole solution, saying sufficiently powerful models could evade shutdown attempts, and instead advocated a “Swiss cheese model” of layered defenses.

He highlighted Anthropic's biosecurity progress, noting that its models are so tightly restricted that biology students often mock the company. He also mentioned that the company recently made progress in protein binding, an early step in drug discovery. Reflecting on his personal experiences—his father died of hepatitis C and he survived early-stage cancer—Amodei stressed the dual responsibility of advancing the technology while managing its risks. He said the dangers are real and the industry should not hide them from the public, but users should have a voice in how AI is deployed.