OpenAI will let the US government inspect its most advanced AI models before public release. The arrangement is voluntary and limited to a 30-day review window. This move follows a recent executive order signed by President Trump.

The order targets “frontier” AI systems with potential national security risks. It allows agencies to assess cybersecurity vulnerabilities and misuse potential before these powerful models go live. The goal is to catch threats early without stifling innovation.

Earlier drafts proposed a 90-day review period, but industry pushback trimmed it to 30 days. The process is not a licensing or approval system—companies can skip it without legal penalty. Instead, it creates a channel for confidential government feedback.

OpenAI’s CEO Sam Altman supports the order but wants oversight handled by civilian scientific regulators, not the National Security Agency. The company favors the Commerce Department’s Center for AI Standards and Innovation to lead testing. This divergence highlights ongoing debate over who should police AI safety.

The executive order also mandates the creation of a classified benchmarking system to evaluate AI cyber capabilities. Agencies have 60 days to build an AI cybersecurity clearinghouse to share findings across government. The White House aims to strike a balance—boosting security without erecting hard regulatory barriers.

For OpenAI and competitors like Google and Anthropic, this means a new step in their release pipeline. They can provide early access to models, receive government input on risks, and still launch at their own pace. Skipping the review won’t break the rules but could raise eyebrows.

Industry insiders worry the order still falls short of the rigorous safeguards needed for truly dangerous AI. Some lawmakers call it a “wild west” policy, granting too much freedom to developers while leaving security gaps.

Still, the government’s quiet re-entry into pre-launch model scrutiny signals a shift. AI is no longer just a tech problem. It’s a national security issue requiring coordination across agencies and private labs.

OpenAI’s upcoming talks with White House officials and lawmakers will test how this voluntary framework evolves. The company plans to push for clearer criteria defining when a model crosses the “frontier” threshold. That line will determine which systems get extra scrutiny.

This new chapter in AI regulation blends caution with pragmatism. The US wants to lead in AI, not choke it under red tape. But the question remains—can voluntary government review keep pace with AI’s breakneck evolution?