BIS Framework for AI Diffusion — Interim Final Rule (RESCINDED)

The BIS Framework for Artificial Intelligence Diffusion was an Interim Final Rule issued by the Bureau of Industry and Security (BIS) (BIS) of the U.S. Department of Commerce on January 13, 2025. It revised the Export Administration Regulations (EAR) to add a three-tier country framework for advanced-computing exports metered by a Total Processing Performance (TPP) compute metric, and created a new Export Control Classification Number, 4E091, for closed-weight dual-use AI model weights trained with at least 10²⁶ operations. BIS rescinded the rule on May 13, 2025, two days before its May 15, 2025 compliance date. It is retained here as a detailed public articulation of a compute-threshold and country-tier AI export-control framework, and continues to be cited in debates on AI diffusion strategy.

Status and timeline

The rule was signed and took effect on January 13, 2025, and was published in the Federal Register on January 15, 2025 (2025-00636; 90 FR 4544). Its compliance date was set for May 15, 2025. BIS rescinded the rule on May 13, 2025, two days before that compliance date, which was never reached.

The pre-existing October 7, 2022 advanced-computing rules and their October 17, 2023 updates, which control advanced chips to China and Country Group D:5, were not affected by the Diffusion Rule and remained in force after the rescission. The America's AI Action Plan is the Trump successor export-control strategy.

After rescinding the rule, BIS issued three replacement documents in May 2025: an AI Model Training Policy; GP10, guidance warning against use of Chinese-made advanced chips, notably the Huawei Ascend line; and Diversion Guidance tightening controls against end-use diversion.

Scope and definitions

The rule revised the Export Administration Regulations across four areas:

• Advanced-computing ICs (ECCN 3A090) — revised controls on advanced-computing integrated circuits.
• New ECCN 4E091 — controls on advanced closed-weight dual-use AI model weights.
• Three-tier country framework — country tiers metered by Total Processing Performance (TPP).
• New license exceptions — LPP, UVEU, and NVEU, along with expanded Data Center Validated End User pathways.

Key provisions

Three-tier country framework

The rule sorted destinations into three tiers, each with distinct treatment for advanced chips and controlled model weights.

Total Processing Performance (TPP) compute metric

Compute quantities under the framework were metered using a Total Processing Performance (TPP) metric. Under License Exception LPP, a Tier 2 entity could receive up to 26.9M TPP per year without individual licensing. The Tier 2 per-country allocation was set at 790B TPP aggregate over 2025–2027. NVEU allocations were calibrated so that NVEU clusters would trail frontier clusters by "approximately 12 months, or one generation."

License exceptions

• UVEU (Universal Validated End User) — available only to Tier 1-headquartered entities. No more than 25% of an entity's AI compute could be located outside Tier 1, and no more than 7% in any single Tier 2 country. US-headquartered providers were required to keep at least 50% of compute domestic.
• NVEU (National Validated End User) — per-country authorizations for Tier 2 data-center operators that provided security compliance documentation.
• LPP (Low Processing Performance) — a streamlined exception for low-volume transactions.

ECCN 4E091 — closed-weight AI model weights

The new ECCN 4E091 covered closed-weight dual-use AI models trained with at least 10²⁶ operations. Open-weight models were not controlled regardless of compute, and closed-weight models trained below 10²⁶ operations were not controlled. Exports of controlled weights from Tier 1 to Tier 2 were permitted under security conditions, while exports to Tier 3 were prohibited. The rule was the first U.S. attempt to regulate model-weight exports.

Relation to predecessor and successor controls

The Diffusion Rule augmented rather than replaced the October 7, 2022 advanced-computing rules and their October 17, 2023 updates, which remained operative after the rescission. The framework is positioned alongside other instruments as follows.

The America's AI Action Plan replaces tiered caps with a more bespoke allied-partnership approach.

Reactions and debates

Several aspects of the rule drew debate, and one of them is cited as a factor in its rescission.

Using compute as a unit of control was new, and TPP metering was untested in enforcement practice before the rescission. The closed-weight carve-out was contested: critics argued that controlling only closed-weight models effectively subsidized open-weight diffusion to adversaries, while proponents argued that open weights were already uncontrollable. Israel, Poland, Saudi Arabia, and others publicly objected to the per-country caps applied to Tier 2 destinations, and this diplomatic backlash is cited as having accelerated the rescission. The broader question of whether and how to control model-weight exports, which the rule was the first to attempt, remained unresolved after the rescission.

The America's AI Action Plan takes a different approach to the diffusion mechanism while sharing the China-denial objective.

Relationships

• supersedes: augmented (did not replace) the October 2022 / October 2023 advanced-computing rules, which remain in force
• superseded-by: America's AI Action Plan — Trump successor strategy
• depends-on: Bureau of Industry and Security (BIS) — issuing agency
• related: Executive Order 14110 — Safe, Secure, and Trustworthy AI — conceptual predecessor on compute reporting
• related: Export Controls (AI), Compute Governance, AI Diffusion, US-China AI Competition: Different Races, Different Metrics
• related: Nvidia & TSMC — AI Compute Infrastructure, DeepSeek
• contradicts: America's AI Action Plan on diffusion mechanism; aligned on China-denial objective

Sources

• BIS AI Diffusion Rule source summary