Anthropic chief executive Dario Amodei has argued that the pace of capability development in advanced artificial intelligence models should be reduced to give safety and oversight work more time. Amodei published an essay titled “We Must Pace the Frontier” on September 12, 2026. Reuters, The Washington Post, The Guardian and Axios reported on the proposal the same day.
The central proposal: give safety work time to catch up
Amodei’s main position is not that artificial intelligence research should stop completely. He argues instead that the introduction of new capabilities should not move so quickly that risk-reduction work falls behind. Progress could remain rapid, according to the essay, while allowing additional time for safety evaluations, outside oversight and coordination among institutions.
The Anthropic executive organises his approach around three broad areas: more extensive testing by companies, stronger access for independent evaluators and common safety rules among leading laboratories. International cooperation is also presented as part of the framework, because companies operate across borders and model development has consequences beyond a single market.
Emphasis on independent evaluation
Amodei argues that internal testing alone may not be sufficient and that outside specialists should be able to conduct continuing assessments. The proposal matters because it could reduce reliance on safety conclusions produced by only one company. The scope of such access, the legal regimes governing it and protections for commercial information have not yet become a detailed common standard.
Reuters reported that Amodei believes a gap is emerging between the growth of model capabilities and the ability to prevent harmful outcomes. The Washington Post and The Guardian described the essay as part of a wider debate among major technology leaders about development speed. Axios focused on the potential effect of the proposals on industry coordination and regulation.
A company proposal is not an industry agreement
The essay represents Amodei’s views and Anthropic’s stated approach. It is not a binding agreement accepted by every artificial intelligence company, and it is not a new rule adopted by governments. Expressions of support from other executives do not by themselves establish that a detailed, enforceable implementation plan exists.
That distinction is important. The confirmed development is that Amodei has publicly proposed slower capability growth, independent evaluation and coordination across the industry. No shared timetable, measurement system or enforcement mechanism has been announced. Claims about what the entire sector will do should therefore wait for formal commitments from the organisations involved.
Why the call came now
The proposal arrives as artificial intelligence systems are able to complete a wider range of tasks and companies are releasing new models at shorter intervals. Amodei contends that safety methods may struggle to advance at the same pace as model capabilities. This is the executive’s assessment; independent researchers differ over the scale and timing of the risks.
His essay also acknowledges that competitive pressure can make it difficult for one company to slow down on its own. It therefore proposes comparable expectations across multiple laboratories and, when necessary, participation by public authorities. Any such coordination would also need to address competition law, national security and international technology policy.
What to watch next
The practical effect of the proposal will depend on the access Anthropic provides to independent evaluators, responses from other laboratories and any action by regulators. For now, the text is a policy appeal. Anthropic did not announce the cancellation of a specific model, and no industry-wide pause has formally entered into force.
The September 12 development shows that artificial intelligence safety is increasingly being discussed not only through technical testing but also through development speed and institutional governance. Whether the proposal changes company road maps will become clearer through subsequent announcements and the creation of workable oversight procedures.
