Business, technology, markets and policy analysis from the United States
Petition · American Commerce Review

Pause the next leap in frontier AI until independent safety tests can keep up

Developers should not train substantially more capable frontier systems faster than independent evaluation and emergency safeguards can be established.

Petition targetCongress, federal agencies and frontier AI developers
Campaign goal25,000 signatures

Every signature adds visible support to a clear, practical request.

Why we are asking

The petition

We are calling for a temporary, capability-based pause on training the next generation of substantially more powerful frontier AI systems when independent evaluators cannot adequately test their dangerous capabilities before deployment. The pause should end when defined evaluation, incident-response and security requirements can be met.

Why this is different from stopping AI

This is not a request to shut down AI products, university research, medical applications, open scientific work or ordinary commercial deployment. The focus is the small number of projects pushing the frontier of general-purpose capability at very large scale, particularly where developers themselves cannot confidently rule out severe misuse or control risks.

Why supporters want a brake

The 2026 International AI Safety Report describes growing evidence of AI use in cyber operations, heightened concern about biological misuse and increasing difficulty conducting reliable pre-deployment evaluations. If capability advances faster than the tools used to test it, waiting for a serious incident may be an unusually expensive way to discover where the boundary was.

Why the objection is serious

A unilateral American pause could move investment, talent and strategic capability overseas. Safety thresholds can also become incumbent-friendly regulation that only the largest companies can afford. Any pacing mechanism therefore needs precise triggers, independent oversight, international coordination where feasible and a fixed process for lifting restrictions.

What we are asking for

Define measurable frontier thresholds, require independent evaluation above them, establish emergency incident-reporting and model-security standards, and create a temporary stop mechanism only when a developer cannot demonstrate that those safeguards are ready. The burden should be proportional to capability, not imposed on every company that uses AI.

Read both sides of the debate

Before signing, read the strongest evidence-based arguments for and against this proposal.

Add your name

A petition works by making support visible. Sign above, then share this page with people who care about the issue.

Build momentum

Share this petition

Help more people see the campaign.

For Instagram, copy the link and share it in a story, message or bio.

American Commerce Review