Anthropic is urging governments to move beyond transparency rules and adopt a more formal regulatory system for frontier AI, arguing that the technology has advanced far faster than public oversight.

In a policy essay released by Chief Executive Dario Amodei, the company says the last few years have shown that AI is no longer a distant possibility but a strategic technology with immediate security and safety implications. Anthropic says recent model capabilities, especially in cybersecurity, have made it clear that existing policy tools are not enough.

The company frames the challenge as a mismatch in speed. AI systems are improving rapidly, while legislation and regulatory institutions typically move slowly. That gap, Anthropic argues, has left policymakers trying to catch up after the technology has already begun reshaping the economic and security landscape.

## From transparency to binding rules

Anthropic has previously backed transparency measures that require AI developers to disclose safety practices, testing methods and serious incidents. The company says those steps were useful when the biggest risks were still uncertain. But it now believes the policy response must be stronger.

The new proposal recommends a regulatory model closer to the way governments oversee aviation or pharmaceuticals. Under that approach, frontier models would face mandatory testing and auditing before release, with the possibility that deployment could be delayed or blocked if a model is judged too risky.

Anthropic says any such system should focus on four concrete risk areas: cybersecurity, biological weapons, loss of control over AI systems and automated research and development that could intensify those threats. It also says advanced AI developers should be required to maintain strong security protections around model weights, carry out regular red teaming and penetration testing, and report significant safety incidents quickly.

The company says testing could be performed either by a government agency or by private evaluators authorized and monitored by the government. It describes this as a possible “regulatory markets” model.

## Security concerns are central

The essay points to recent frontier-model behavior as evidence that AI risks are no longer theoretical. Anthropic says one of its recent models exposed weaknesses in the global cybersecurity environment, underscoring the possibility that advanced systems could be misused against financial institutions, critical infrastructure or national security targets.

The company argues that these incidents demonstrate why policymakers need tools that can respond to real-world risks, not just hypothetical ones. It says the first wave of transparency laws should be seen as preparation for more targeted rules based on what researchers and regulators learn from model behavior.

## Broader policy implications

Amodei’s essay also looks beyond regulation, suggesting that powerful AI could complicate economic policy by accelerating growth while also widening inequality and causing long-lasting labor disruption. But the strongest near-term focus is on safety governance.

Anthropic says governments should avoid waiting for every future danger to become fully understood before acting. Instead, it wants policy designed around the risks already emerging, with room to tighten rules if models become even more capable.

The company said it is also releasing a legislative proposal on frontier model testing and a separate policy framework on job displacement, and that it plans to support both financially.

While Anthropic says it welcomes recent government movement toward greater AI oversight, it argues that the pace of policy still trails the pace of model development. Its message is that regulators should prepare now for a technology that is becoming more powerful, more widely deployed and, in the company’s view, too consequential to leave largely to voluntary safeguards.