ai risk today logo svg bl font
Menu
SUBSCRIBE
SUBSCRIBE
ai risk today logo svg bl font
Menu

airisktoday

About airisktoday

https://airisktoday.wpenginepowered.com

Posts by :

EU Delays EU AI Act, Digital Omnibus package

EU Moves To Delay Full Enforcement Of AI Act Under New Digital Omnibus Package

, , ,

The European Commission proposed delaying several core compliance deadlines under the EU AI Act as part of a broader “Digital Omnibus” package intended to align and simplify EU digital regulations. The measures would postpone the Act’s strictest requirements, including those covering high-risk AI systems, by more than a year beyond the original schedule.

Key Changes to Enforcement Timelines

Under the EU AI Act, adopted in 2024, high-risk systems, such as those used in credit scoring, biometric identification, recruitment, and medical devices, were required to comply by August 2026 (Annex III) and by August 2027 for systems embedded in regulated products (Annex I).

The Commission’s new proposal revises these deadlines to:

  • Annex III: High-risk AI systems, December 2027.
  • Annex I: Regulated products, August 2028. This assumes harmonized standards are not yet in place.

Once standards or official guidelines are available, obligations may take effect six months after publication for Annex III systems and 12 months for Annex I.

 

The Digital Omnibus package also introduces administrative streamlining measures to reduce overlap across the AI Act, GDPR, and other digital policy frameworks.

Rationale for the Delay

The Commission stated that additional time is needed to finalize technical standards, develop conformity-assessment mechanisms, and ensure that national authorities can consistently supervise and enforce the law. Officials warned that prematurely enforcing high-risk requirements without these elements could result in fragmented national implementation and non-compliance.

Industry groups support the extension, arguing that the original timeline posed operational challenges for healthcare, finance, manufacturing, and other sectors deploying high-risk AI. Companies noted that documentation, model testing, and oversight requirements require significant preparation.

Concerns Raised by Lawmakers and Civil Society

Consumer protection and digital rights organizations expressed concern that the delay weakens core protections intended to prevent the deployment of biased or unsafe AI. Critics argue that slower enforcement may disproportionately benefit large technology providers, who can scale AI systems in sensitive domains before full obligations take effect.

Some EU lawmakers also questioned the use of an omnibus legislative package, noting that bundling multiple regulatory adjustments could limit scrutiny typically applied to major policy changes.

Implications for Businesses

The Commission emphasized that the postponement should not be viewed as an opportunity to defer compliance planning. Pending standards, potential early activation of obligations, and ongoing rulemaking mean timelines could still shift.

Organizations developing or deploying high-risk AI are expected to continue preparations, including documentation, risk assessment procedures, and data governance controls. The Commission said it aims to simplify compliance for small and medium-sized enterprises, though details are not finalized.

Next Steps

The proposal must first be approved by both the European Parliament and the Council. Several Members of Parliament have already signaled caution about reopening the newly enacted AI Act. The final timeline will depend on legislative negotiations and the pace at which technical standards and supervisory frameworks are completed.

If adopted, the revised schedule would represent a significant adjustment to Europe’s flagship AI regulatory framework, extending the implementation runway while maintaining the option of earlier enforcement where standards are ready.

U.S. AI Regulations allowing NVIDIA to sell B30A chips to China.

New Analysis Warns: Exporting NVIDIA’s Blackwell B30A Chips To China Could Erode U.S. AI Advantage

, , , ,

A new report from the Institute for Free Policy (IFP) evaluated whether the United States should allow exports of NVIDIA’s B30A Blackwell-architecture AI chips, a scaled-down version of NVIDIA’s flagship B300, designed for the Chinese market. The report indicates that permitting these exports could significantly reduce the U.S. lead in AI compute and alter the trajectory of global AI development.

A Downgraded Chip With High Strategic Impact

According to IFP, the B30A is expected to deliver approximately half the performance at half the cost, making it a cost-effective option for Chinese AI companies seeking large-scale compute.

Despite its downgraded profile, the chip is projected to exceed U.S. export-control performance thresholds by a wide margin. The report notes that allowing the B30A into the Chinese market would represent a substantial departure from the current strategy, which restricts high-end AI compute from reaching geopolitical competitors.

Modeling Shows a Sharp Decline in U.S. Compute Advantage

IFP modeled multiple scenarios through 2026. Under existing export restrictions, the United States is projected to maintain a 31× advantage over China in frontier AI compute. This assumes zero importing of the B30A into China and effective enforcement of smuggling laws. That dynamic changes markedly if the Blackwell chips are approved for export.

In a limited-export scenario, the U.S. advantage shrinks to under 4×. In a more permissive scenario, one that includes selling a greater supply of chips, China could reach parity or even gain a slight edge by 2026.

The report emphasizes that compute capacity is the primary driver of AI progress. A narrowing gap in available compute could accelerate China’s ability to train large-scale frontier models and compete directly with U.S. capabilities.

The argument supporting the export of downgraded chips is that doing so might slow China’s domestic semiconductor efforts, as NVIDIA and other US chipmakers will take market share from Chinese companies such as Huawei. The IFP analysis challenges this view for two reasons –

  1. Chinese manufacturers, most notably Huawei, cannot produce chips to meet demand anyway. This is, in part, due to U.S. and allied export controls on semiconductor manufacturing equipment (SME) that limit production.
  2. China is still driving towards semiconductor independence. The government is pushing state-owned companies to adopt domestic chips while simultaneously limiting AI chip imports. 

Policy Recommendations Focus on Manufacturing Controls

Rather than approve exports of the B30A, IFP recommends strengthening controls on key chokepoints within the semiconductor supply chain:

  • Semiconductor manufacturing equipment: Tighten restrictions on tools such as DUV immersion lithography machines and close circumvention pathways through third countries.
  • High-bandwidth memory (HBM): Increase enforcement and due diligence against HBM suppliers to ensure compliance with export rules.
  • Component-level oversight: Enhance monitoring of chip assembly, packaging, and advanced interconnect technologies that could enable China to scale production.

While NVIDIA and other American companies may see commercial upside in expanded sales, the report suggests that the long-term strategic implications outweigh near-term revenue gains. Allowing Blackwell chips into the Chinese market could accelerate model development in ways that current U.S. policy is designed to prevent.