Key Takeaways

  • The transition from Executive Order 14110 to the 2025 Executive Order 14179 marks a formal federal shift toward removing regulatory barriers to artificial intelligence development.
  • Despite the deregulatory focus, voluntary pre-release reviews continue, with government evaluators receiving up to 30 days of access to OpenAI’s upcoming Astra model.
  • While over 55% of organizations are experimenting with generative AI, fewer than 25% have integrated formal risk frameworks into their product release gates.

The landscape of U.S. artificial intelligence oversight is undergoing a fundamental policy shift, replacing broad mandates with targeted, voluntary industry cooperation. The Biden administration’s Executive Order 14110, issued in October 2023 to mandate reporting and safety testing for powerful models, was formally revoked by the Trump administration. In its place, Executive Order 14179, issued in 2025, directs federal agencies to prioritize removing barriers to American AI leadership and reassessing prior regulatory actions.

A more deregulatory federal posture changes the mechanics of government involvement rather than eliminating it entirely. Instead of comprehensive compliance rules for model development, Washington favors voluntary evaluations and industry cooperation. OpenAI CEO Sam Altman recently confirmed that the Trump administration conducted a voluntary review of OpenAI’s forthcoming Astra model, giving government evaluators access to the system for up to 30 days before its public release. Altman described the testing to Axios as a “productive process” and emphasized that close engagement with safety institutes in the United States and the United Kingdom will become increasingly critical as models advance in capability.

This dynamic highlights an ongoing debate over frontier model oversight. While OpenAI participates voluntarily, policymakers must determine if comparable pre-release reviews will apply consistently across the industry to systems like Anthropic’s Claude models and Google DeepMind’s Gemini. Relying entirely on voluntary cooperation leaves open questions about how emerging or less visible companies will handle rigorous safety testing before deployment.

For enterprise technology leaders, navigating this evolving environment requires internal safeguards, particularly as external mandates fluctuate. According to 2024 Gartner data, over 55% of organizations are experimenting with generative AI, with regulation and governance cited as top barriers to scaled deployment. Furthermore, a 2024 McKinsey report indicates that while more than 40% of enterprises using generative AI have implemented formal AI risk frameworks, fewer than 25% actually integrate those frameworks into their product release gates.

Global regulatory pressure also complicates the compliance landscape for multinational enterprises. While international reporting from outlets like Reuters often highlights friction over global AI rules, OECD data from 2024 confirms that G20 governments have accelerated their AI governance measures post-2023, with most establishing national AI strategies and baseline risk guidelines. The commercial and geopolitical stakes draw intense interest from key industry players, including leadership at hardware providers like Nvidia and model developers like Meta, alongside prominent technologists such as Elon Musk and David Sacks, who are closely watching how fragmented global rules will impact deployment.

Even without strict federal mandates in the U.S., resources like the NIST AI Risk Management Framework remain highly relevant. The 2023 framework provides a technology-neutral baseline for enterprises deciding how to document model risks, assign accountability, and test systems before deployment. While the federal government may decline to impose a new licensing regime, corporate boards, customers, insurers, and international regulators continue to demand robust evidence of testing, ensuring that formal assurance and evaluation regimes remain critical for enterprise AI adoption.