Why Anthropic CEO Dario Amodei wants AI regulated like aviation and pharma

Anthropic chief Dario Amodei says advanced AI now poses public safety and national security risks, calling for mandatory testing and government oversight before deployment

Anthropic chief executive officer (CEO) Dario Amodei has argued that advanced artificial intelligence (AI) systems should be regulated like aircraft and pharmaceutical products, warning that frontier AI models now pose "real public safety" and "national security risks".

 

In his essay Policy on the AI Exponential, Amodei argued that while AI is advancing at extraordinary speed, governments and legislatures are moving much more slowly. If this trend continues for another year or two, he argues, the world could see what he calls "Powerful AI", equivalent to "a country of geniuses in a datacentre". 

 

Why Amodei believes regulation is necessary

 

Amodei said evidence of both AI's power and its risks has become impossible to ignore. He cites Anthropic's Claude Mythos Preview as a key example, arguing that frontier AI models have demonstrated capabilities that could pose significant cybersecurity risks, potentially threatening financial systems, critical infrastructure and national security.

 

 

He warned that cybersecurity risks may only be the beginning. Biological threats and serious AI autonomy risks could follow as systems become more capable.

 

According to Amodei, the AI industry initially focused on transparency, but transparency alone is no longer sufficient. Current AI systems resemble technologies such as cars, aircraft and medicines—essential to modern society but potentially dangerous if poorly designed or deployed. Over time, he argued, powerful AI could even begin to resemble nuclear materials, requiring far stricter oversight if governments fail to keep pace. 

A regulatory model similar to aviation

 

Amodei proposed creating an AI regulatory framework modelled on agencies such as the US Federal Aviation Administration (FAA).

 

His proposal includes:

 

• Mandatory testing for advanced AI models above a compute threshold
• Assess risks in cybersecurity, biosecurity, AI autonomy and automated R&D
• Allow governments to block or delay unsafe models
• Require strong security measures, red-teaming and threat monitoring
• Use government agencies or accredited independent bodies for evaluations

The economic challenge: Growth and inequality

 

The essay also highlights AI's potential impact on labour markets. Amodei says AI could eventually perform most cognitive tasks better than humans, creating unprecedented economic growth while simultaneously increasing inequality.

 

In such a scenario, he argued, the challenge would no longer be generating growth but ensuring that its benefits are widely shared. Significant and lasting job displacement could become an inherent feature of AI-driven economies.

 

To address these concerns, he proposed:

 

• Better measurement and tracking of AI-driven job displacement
• Pro-employment incentives aimed at slowing or reducing workforce disruption
• Long-term income support mechanisms if labour demand declines permanently
• Exploration of Long-term maroeconomic support
• Universal capital accounts as an additional tool for distributing economic gains

 

Amodei also addressed concerns over data centres and energy demand, arguing that AI companies should bear the cost of any resulting increases in electricity prices.

 

Protecting democracy and civil liberties

 

Amodei warned that powerful AI could become an unprecedented tool for authoritarian governments if appropriate safeguards are not established.

 

He argued that AI-enabled surveillance could allow governments to analyse vast amounts of public information and infer highly personal details about citizens, capabilities that existing civil liberties frameworks were never designed to address. Similarly, future autonomous weapons systems could enable governments to exercise power with reduced human oversight.

 

To safeguard democratic institutions, Amodei proposed:

 

• Clear accountability rules for fully autonomous weapons
• A ban on domestic deployment of autonomous weapons systems
• Closing data broker and bulk data collection loopholes
• Ensuring citizens have access to AI-based assistance

 

He also argued that AI should not be entrusted entirely to either governments or corporations, calling instead for checks and balances on both.

 

The geopolitical race for AI

 

According to Amodei, AI will become a major source of geopolitical power. A nation possessing powerful AI could hold an advantage comparable to a modern military confronting a medieval army. He advocated the formation of a coalition of democratic nations built around shared AI values.

 

Key elements of such a coalition would include:

 

• Free sharing of advanced chips and semiconductor manufacturing equipment
• Coordinated management of AI supply chains
• Joint efforts to address AI-related risks
• Shared access to AI's economic and technological benefits
• Mutual defence against adversarial AI systems
• Rejection of AI-enabled authoritarian repression
• Macroeconomic cooperation among member states

 

Preparing for an explosion of AI-driven innovation

 

Beyond AI itself, Amodei argued that existing regulatory systems are unprepared for the surge of innovation AI could unleash across industries.

 

He highlighted biomedical research as a key example. AI could dramatically accelerate drug discovery and medical innovation, but regulatory approval systems remain designed for a much slower pace of scientific progress. Current approval processes at agencies such as the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) can take seven to eight years.

 

Without reform, Amodei warned, AI-driven innovation could overwhelm existing regulatory structures. He suggested regulators begin developing standards now for AI-assisted research methods so that proven innovations can be adopted quickly once validated.

 

Why Anthropic is pushing for stronger AI oversight

 

Keeping in line with the concerns outlined in his essay, Amodei also highlighted steps Anthropic has already taken, as well as measures it plans to support in the future.

 

He said the company is backing proposals on frontier AI model testing and job displacement, while supporting transparency legislation in California, New York and Illinois. According to Amodei, the key challenge is ensuring that governments and institutions can adapt quickly enough to manage AI's growing risks while maximising its benefits.