Anthropic CEO Dario Amodei released an essay today urging governments to move quickly to establish binding regulation for artificial intelligence, saying that current voluntary transparency measures cannot sufficiently manage the risks posed by the technology.
Amodei recommends a regulatory framework modeled on the Federal Aviation Administration - an approach that would require mandatory third-party testing of powerful AI systems and grant government authorities the power to prevent unsafe models from operating. The proposal follows recent cybersecurity disruptions linked to the Claude Mythos Preview system.
Five policy areas for immediate attention
In his essay, Amodei identifies five domains where he says policy action is urgent.
- Regulation and public safety - Amodei argues that AI presents risks on a scale comparable to aviation or nuclear technology. He calls for compute-heavy models to undergo compulsory testing that assesses vulnerabilities related to cybersecurity, biological weapons, potential loss of control, and risks from automated research and development.
- Economic policy - Cautioning that AI could lead to substantial job displacement, Amodei proposes a mix of near-term and long-term responses. Short-term measures he recommends include wage insurance, retention tax incentives, and retraining grants. For the longer term, he suggests exploring universal basic income or universal capital accounts, to be financed by taxing the economic gains generated by AI.
- Scientific advancement - Amodei suggests that regulatory bodies such as the Food and Drug Administration should update approval pathways to accept AI-driven methods, including simulations and synthetic control arms, with the goal of shortening the current seven to eight year timelines for bringing new treatments to market.
- Civil liberties - He calls for a ban on the domestic use of autonomous weapons and wants legal loopholes closed that permit governments to buy bulk data from private firms. He also recommends that citizens be guaranteed access to AI systems matching the capability of those used by governments in legal settings.
- Geopolitics - Amodei proposes that democratic countries form a selective coalition to manage the AI supply chain - encompassing chips and manufacturing equipment - coordinate mutual defense, and limit economic and scientific sharing to allied nations.
Throughout the essay, Amodei emphasizes the speed of AI progress. He warns the technology is advancing exponentially and suggests it could reach the equivalent of a "country of geniuses in a datacenter" within a few years.
The proposals aim to pair technical safeguards - such as third-party evaluation and the power to block unsafe systems - with economic and legal measures intended to mitigate societal disruption and protect civil rights. Amodei frames these steps as responses to both recent operational incidents and the broader trajectory of AI capability growth.