Dario Amodei draws a parallel between policymakers and Treebeard from Lord of the Rings—the slow-moving tree who cannot keep pace with urgent threats. AI capabilities are advancing exponentially, with models evolving from basic code generation to systems like Claude Mythos Preview that can disrupt cybersecurity and pose national security risks.

For years, responsible AI advocates including Anthropic pushed transparency measures like California's SB 53 and similar legislation in New York and Illinois. This approach preserved optionality while policymakers caught up to AI's trajectory. The logic: gather data on emerging risks before crafting binding regulations.

Now Amodei argues that Claude Mythos Preview has proven AI's strategic significance beyond doubt. He advocates modeling AI oversight after the Federal Aviation Administration, requiring technical testing and audits for frontier models before deployment. The cybersecurity vulnerabilities exposed demand immediate regulatory action, not continued observation.

The mismatch between AI's exponential curve and government's deliberate pace creates dangerous blind spots. While policymakers show increased willingness to act, their responses lag at least a year behind current capabilities. Binding regulation modeled on cars, airplanes, and drugs is the only viable path forward.