Anthropic CEO Dario Amodei has called for mandatory testing and government oversight of advanced artificial intelligence models, highlighting that frontier systems now pose sufficiently serious cybersecurity, biological and national-security risks to warrant regulation similar to that applied to aircraft and pharmaceuticals.
In a policy essay published on Thursday and shared by him on X, Amodei said the AI industry had moved beyond a phase where transparency and voluntary disclosures were adequate safeguards. Instead, governments should have the authority to evaluate and, where necessary, block the deployment of advanced AI systems that present unacceptable risks.
“Now the risks are clearly here,” Amodei wrote, adding that it was “time to go beyond transparency to more serious and binding regulation of AI”.
Shift From Voluntary Disclosure
The proposal marks a notable evolution in the debate around AI governance. Over the past two years, several leading AI firms, including Anthropic, have supported measures requiring developers to disclose safety practices, testing procedures and major incidents. Amodei wrote that while such measures helped policymakers better understand emerging risks, the rapid improvement in AI capabilities now requires a more interventionist approach.
Under Anthropic’s proposed framework, AI models above a certain computing threshold would undergo mandatory testing by qualified third parties. The evaluations would focus on four areas: cybersecurity risks, biological threats, loss of control of AI systems, and the ability of AI to accelerate research that could amplify those risks.
The government, according to the proposal, should have the power to block or deter deployment of systems deemed unsafe, based on the outcome of those assessments. AI developers would also be required to maintain stronger security standards, conduct regular red-team testing and promptly report serious safety incidents.
Amodei compared advanced AI systems to technologies such as commercial aviation and medicines, where independent testing and regulatory oversight are considered prerequisites for public deployment. He has pointed that existing policy approaches were struggling to keep pace with the speed of AI development.
The Anthropic CEO said that AI was advancing far faster than traditional policy-making cycles. While governments often take years to legislate, he said that AI systems had progressed within a few years from limited capabilities to performing increasingly complex tasks across software development, science, mathematics and other knowledge-intensive domains.
Economic Disruption
Beyond regulation, the essay also addressed the economic implications of increasingly capable AI systems. Amodei warned that AI could eventually become a substitute for a broad range of cognitive work, creating the possibility of sustained labour-market disruption even as overall economic growth accelerates.
He urged governments to expand efforts to measure AI’s impact on employment and consider interventions such as wage insurance, worker-retention incentives, workforce training programmes and, if required, longer-term income support mechanisms.
“The key challenge in such a world won’t be incentivising growth, but finding a way for everyone to share in the benefits,” he wrote.
The essay also framed AI as a strategic technology with implications extending beyond commercial competition. Amodei said that countries leading in advanced AI could gain significant economic and military advantages, making AI policy increasingly intertwined with national-security considerations.
He backed tighter controls on the supply chain supporting advanced AI development, including export restrictions on leading-edge semiconductors and chipmaking equipment. Amodei also called for closer coordination among democratic nations on AI regulation, security standards and deployment policies.
The recommendations come amid growing scrutiny of frontier AI systems by governments across the US, Europe and parts of Asia. Regulators have increasingly focused on issues ranging from misinformation and copyright disputes to concerns over cybersecurity, critical infrastructure and the potential impact of AI on employment.
While much of the global policy discussion to date has centred on transparency, disclosure requirements and voluntary commitments from technology companies, Amodei’s essay signals a broader shift within parts of the industry towards formal safety certification and regulatory oversight of the most advanced AI models.
