Amazon Calls for AI Release Safeguards Without Backing a Slowdown

Amazon has entered the AI safety debate by saying models should be released only when they are safe to use, after rigorous testing and strong safeguards. It also called for industry cooperation with government on protections, but did not endorse a wider slowdown in AI development.

A release test, not a speed limit

Amazon’s position centers on the decision to put a particular model in users’ hands. An Amazon spokesperson said the company does not see AI progress and safety as competing goals. The standard it described is readiness for use: testing and safeguards should come before release, rather than a general call to restrict development across the industry.

Models should be released when they’re ready and safe to use, which comes from rigorous testing and strong safeguards.

Amazon spokesperson, speaking to Reuters

Coordination is the unresolved part

The company said there are risks if AI progress does not happen collectively and argued that industry should work with governments on the right protections. That puts Amazon behind coordination in principle. But it did not lay out a joint testing system, a common release threshold, or specific safeguards it would adopt alongside other labs.

A debate already split over pace

Amazon spoke after Anthropic CEO Dario Amodei called for the industry to slow development of more powerful AI models. OpenAI, xAI, Google DeepMind and Microsoft later supported a more measured approach, according to Reuters. Amazon did not say whether it would take comparable safety measures.

The positions now on the table

• Amodei called for independent evaluators inside leading AI companies and government action to slow the race toward more powerful models.
• OpenAI CEO Sam Altman supported pacing frontier AI development, while clarifying that pacing does not mean stopping.
• Nvidia and Meta argued that companies should assess and control their own AI risks.

What Amazon’s position leaves open

Amazon has aligned itself with the idea that safety checks should precede model releases and that governments belong in the discussion. Its statement does not settle the harder question raised by the slowdown debate: whether labs should agree on shared constraints before any one company decides a model is ready.