Anthropic CEO Proposes AI Indexed Development Policy, Calls for Mandatory Regulation and Job Security

Anthropic founder and CEO Dario Amodei published a lengthy article in June 2026 titled "Artificial Intelligence Indexed Development Policy," arguing that AI is developing exponentially, far outpacing policy responses, and that risks such as cybersecurity, bioweapons, and AI loss of control are imminent. The article proposes five governance directions: establishing mandatory third-party testing and government shutdown mechanisms modeled after the FAA; addressing job displacement through taxes and employment subsidies; reforming biomedical regulation to unlock AI's positive value; regulating autonomous weapons and data privacy; and advocating for democratic nations to form alliances to control supply chains. Anthropic has committed $350 million ($200 million for research and $150 million for scholarships) to support related proposals.

Core Background: Exponential AI Development Forces Policy Change

Amodei uses the Ents from "The Lord of the Rings" as a metaphor for slow policy systems: AI went from barely writing code to handling most programming work at leading companies in four years, while congressional legislation takes years.
Scaling laws have over a decade of empirical support; if they continue, "strong AI"—equivalent to a "virtual nation" of 100 million top scientists—could arrive in the next 1-2 years.
Key trigger event: Claude Mythos Preview autonomously discovered thousands of high-risk vulnerabilities, proving that frontier AI already has national-level strategic impact, making cybersecurity risks a reality and accelerating the approach of bioweapon and loss-of-control risks.

Regulatory Framework: Mandatory Third-Party Testing and Government Shutdown Authority

Modeled after the FAA, mandatory regulation applies to companies with computing power exceeding 10^25 FLOPs or AI-related revenue exceeding $500 million/R&D spending exceeding $1 billion.
Third-party testing covers four major risks: cybersecurity, bioweapons, AI loss of control, and accelerated automated R&D.
If an assessment finds unacceptable risks, the government has the authority to prevent, delay, or revoke model deployment, with safeguards against political bias.
Model weights are considered strategic assets; developers must maintain high security standards, conduct regular red team testing, and mandatorily report security incidents.

Employment and Economy: Addressing Persistent Unemployment

Amodei believes AI may break the historical buffering mechanisms of technological revolutions (Jevons paradox and comparative advantage), leading to persistent unemployment.
Three-tier economic framework:
- Data tracking: Expand data collection on AI's employment impact.
- Employment incentives: Wage insurance, employer retention tax incentives, and job training subsidies.
- Long-term support: If necessary, fund universal basic income through taxes on AI companies or capital gains taxes.
Anthropic commits $200 million for policy research and $150 million for skills training scholarships.

Unlocking Positive Value: Reforming Biomedical and Other Regulations

Traditional drug approval processes (7-8 years) cannot keep up with AI-accelerated R&D output; reforms needed:
- Accept AI-generated pharmacodynamic/pharmacokinetic simulations.
- Promote AI-predicted toxicology to reduce animal testing.
- Introduce synthetic control groups to shorten clinical trials.
- Establish innovative fast tracks for cancer, Alzheimer's, etc.

State Power and Civil Liberties: Drawing Boundaries

Regulate autonomous weapons use, close data privacy loopholes, and balance public power with civil liberties.

Geopolitics: Democratic Nations Cooperation

Advocate for democratic nations to form alliances to control core AI supply chains and engage in cross-border risk collaborative governance and joint defense.

Industry Reactions and Significance

Amodei states bluntly that "the era of mild self-regulation is over," breaking Silicon Valley's tacit resistance to regulation.
Emphasizes that public concern is not a "PR problem" but a normal manifestation of democratic oversight.
The article is seen as a signal of a critical window for AI governance, calling for consensus and forward-looking policies.