In 2025, the United States did indeed move forward with new or expanded export controls on advanced AI chips, cloud‐based access to AI compute infrastructure, and AI model weights—but with important caveats and subsequent modifications. In January, the Bureau of Industry and Security (BIS) released a sweeping “AI Diffusion” rule that sought to impose global licensing requirements on exports of high-end AI chips, limit access to large compute clusters for AI training, and introduce a new category for model weights exports. However, by mid-May the same agency announced the rule would be rescinded due to concerns about innovation and diplomatic fallout. Simultaneously, instead of a broad and enduring global ban, the U.S. shifted toward a more targeted architecture of export controls, focusing on specified entities, validated end-users, and particular country tiers—thereby lowering the probability of wholesale controls on model weights or cloud compute across all countries. For 2025 the implied outcome is: yes, new controls were introduced—but their long‐term enforcement in that full form did not persist, and the dominant trend is toward selective measures rather than universal license regimes. As such, while companies dealing in AI hardware, cloud providers and model developers should expect incremental and specific regulatory steps, they should not anticipate a blanket export licensing regime for all AI model weights or cloud compute access being fully operational in 2025.
Source: https://www.bis.gov/press-release/department-commerce-announces-rescission-biden-era-artificial-intelligence-diffusion-rule-strengthens?utm_source=chatgpt.com "Department of Commerce Announces Rescission of Biden- ..."
Why do you think you're right?
Although the Trump administration rescinded the 2025 AI Diffusion Rule before it fully took effect—arguing that it was overly complex and bureaucratic—the evidence suggests that the government did not abandon the intention to regulate exports related to advanced AI. In fact, the Department of Commerce announced that it would issue a “replacement rule,” and in October 2025 it established the *American AI Exports Program*, aligned with an executive order aimed at promoting the export of the “American AI Technology Stack.” This indicates that the administration remains interested in creating a formal, legally binding framework to control the international diffusion of critical AI technologies.
Given that the U.S. government views frontier AI as a strategic asset with national security implications—especially regarding China—it is reasonable to expect that the new rule will include explicit controls on advanced model weights, frontier-class training compute, or cloud access intended for frontier-model development. Moreover, rescinding the previous rule did not eliminate the institutional recognition that frontier AI requires a specific regulatory regime; it merely opened the door for a version that is simpler, more focused, and more aligned with the current administration’s vision.
Therefore, with a program already created, an executive order mandating action, and a regulatory gap the administration has committed to fill, it is highly probable that before July 31, 2026 a new rule—published in the Federal Register or as an update to the EAR—will be implemented that meets the required criteria: explicit controls on frontier model weights, frontier-training compute, or cloud access for advanced AI development, with broad applicability and verifiable enforcement.
Why might you be wrong?
It is also possible that the United States will not implement new or expanded export controls targeting frontier AI model development before July 31, 2026. One reason is political and administrative reluctance. The Trump administration already rescinded the Biden-era AI Diffusion Rule on the grounds that it was too complex and burdensome, and that it risked slowing American innovation. This same position could lead the administration to delay or indefinitely postpone any new binding controls, even if officials continue to express interest in crafting a replacement. Additionally, the regulatory process within the Department of Commerce is inherently slow. Drafting a new rule, completing interagency consultations, conducting national security and economic impact assessments, clearing legal review, and publishing in the Federal Register can take well over a year. It is entirely possible that the process will not be completed in time.
Industry resistance may also slow progress. Companies such as OpenAI, Google, NVIDIA, and major cloud providers could push back against new export controls, arguing that they would create unnecessary compliance burdens, undermine competitiveness, and fail to meaningfully hinder adversaries like China. This kind of lobbying has historically influenced BIS timelines and could lead to delays or a quiet deprioritization of the issue. Another important factor is the difficulty of enforcing export controls on digital assets like model weights or on cloud-based access for AI training. Policymakers may worry that any rule they design would be easy to circumvent, technically infeasible to monitor, or would require intrusive measures that the administration is unwilling to implement.
It is also possible that the administration decides to prioritize diplomatic engagement or market-based strategies instead of binding regulations, opting for voluntary frameworks or international agreements that would not meet the resolution criteria. Internal disagreements within the newly created American AI Exports Program could further complicate matters, especially if agencies cannot agree on definitions, thresholds, or country restrictions. Finally, any proposed rule could face legal challenges from industry or civil liberties organizations, delaying its implementation past the forecast window even if the government begins the process.
In sum, political hesitation, bureaucratic complexity, industry opposition, enforcement challenges, and potential legal disputes all present credible pathways through which the United States might not finalize new or expanded export controls on frontier AI development by July 31, 2026.