Trump AI order skips model-level oversight, IBM endorses the gap

The Trump administration's AI executive order, released this week, focuses on cybersecurity infrastructure but does not require frontier model providers to share technical details about their latest systems. IBM CEO Arvind Krishna endorsed that approach at the Axios AI+NY Summit, arguing for "light regulation" and warning against bureaucracy that would slow deployment.

The absence of model-level disclosure obligations is a structural divergence from the Biden-era approach, which pushed toward pre-release safety reporting for large training runs. The new order fortifies cyber defenses but leaves the model layer largely unregulated. Krishna framed that as the correct posture.

The statement matters because IBM is active in enterprise model deployment, not just infrastructure. The company ships watsonx and integrates third-party models at scale. Krishna's read is that the current provider economics—pricing pressure, inference cost compression, open-weight competition—impose tighter discipline than federal mandates would.

The order does not compel release of training compute totals, eval benchmarks, or safety-test results. It does not require advance notice of new model classes. Providers retain full discretion over what they publish. That is the lightest regulatory envelope among major economies with active AI frameworks.

If the posture holds, the constraint on model providers remains economic, not statutory. Pricing per million tokens, open-weight substitution, and hyperscaler capex cycles set the pace. The federal layer steps back.