Dario Amodei wants an FAA for frontier AI

In The Lord of the Rings, it takes the ancient tree giant Treebeard an entire day just to say hello to another tree, while the Hobbits are already sprinting through the woods. To Anthropic CEO Dario Amodei, that painful mismatch is exactly what is happening between frontier AI and Washington.

Amodei published a sweeping policy essay warning that exponential AI progress is leaving policymakers dangerously behind. His core argument is simple: voluntary transparency pledges are no longer enough, and governments must step in with binding safety rules modeled after the Federal Aviation Administration (FAA).

Alongside the essay, Anthropic released an explicit legislative proposal for frontier model testing, paired with a policy framework to tackle AI-driven job displacement—both backed by promised financial support from the company.

Moving past transparency to mandatory grounding

For the past few years, AI safety advocates largely settled for transparency laws, helping pass bills like SB 53 in California, RAISE in New York, and SB 315 in Illinois. The goal back then was just to peek inside the black box and gather data on risks as they surfaced.

Amodei argues that phase is officially over. The turning point was Claude Mythos Preview, which disrupted the global cybersecurity landscape and proved that frontier models now carry profound national security consequences. If cybersecurity was the first major shockwave, biological risks and autonomy threats are right behind it.

To counter this, Anthropic wants an FAA-style regime. Just as commercial planes cannot fly without rigorous technical audits, frontier models exceeding a specific compute threshold should undergo mandatory third-party safety testing before release.

The checks focus squarely on four catastrophic areas: cybersecurity, biological weapons, loss of control, and automated R&D loops that compound those risks. Crucially, if third-party evaluators find unacceptable danger, the government should have the legal power to block or halt deployment entirely.

The proposal also demands enterprise-grade defenses for model weights, mandatory penetration testing, and immediate reporting of critical incidents.

Preparing for the economic shockwave

Beyond bioweapons and zero-day exploits, Amodei took direct aim at the labor market. If scaling laws hold for another year or two, we could see what he calls a country of geniuses running inside a datacenter.

While that could unleash unprecedented economic acceleration, it also threatens to permanently replace general human cognitive labor. Amodei warns the economy could get trapped in a hypergrowth, hyper-inequality setting where traditional job transition playbooks completely break down.

That explains why Anthropic paired its security proposal with an economic transition framework, signaling that labs building the technology must also help bankroll the societal safety net.

Why it matters

We think Anthropic is making a remarkably calculated move here, effectively drawing the regulatory perimeter before anyone else can do it for them.

For the broader AI ecosystem, calling for mandatory audits and deployment vetoes fundamentally changes the game. Establishing an FAA for code sets up massive compliance hurdles that well-funded frontier giants can absorb, but mid-sized challengers and open-weight efforts might struggle to navigate.

For end users and developers, this shift likely means slower rollouts and heavier guardrails on frontier models, but far less anxiety about catastrophic zero-day exploits or biological misuse slipping into wild deployments.

The open question is whether a slow-moving political apparatus can build an FAA-caliber agency without turning it into a tool of bureaucratic capture or political favoritism. But by handing the leash directly to regulators, Anthropic has made one thing clear: the era of self-policing frontier AI is effectively over.