EU Council and Parliament Reach Deal to Simplify AI Act—Here's What Changed

By Hector Herrera | May 21, 2026 | Government

The European Union's Council and Parliament reached a provisional agreement in May to simplify key provisions of the AI Act — the world's most comprehensive AI regulation — responding to sustained criticism that the original framework was operationally unworkable for startups and mid-size companies. The deal does not abandon the Act's risk-based architecture, but it reduces compliance overhead in targeted areas and clarifies definitional ambiguities that had paralyzed procurement decisions.

The AI Act passed in 2024 as the first comprehensive national or regional AI law in history. Its compliance obligations — particularly for "high-risk" AI systems in employment, education, healthcare, and critical infrastructure — generated immediate and loud pushback from European tech companies who argued that audit requirements, documentation standards, and conformity assessment procedures were sized for large enterprises, not the startups and SMEs most likely to build European AI.

What the agreement does

According to the Council of the European Union, the provisional deal reduces compliance requirements for smaller developers, streamlines certain conformity assessment procedures, and resolves definitional ambiguities about which systems qualify as "high-risk." The full simplified text has not yet been finalized, but the political agreement establishes the direction.

The core risk framework — with four tiers from unacceptable risk (banned systems such as real-time public biometric surveillance) through high-risk, limited-risk, and minimal-risk — remains intact. What changes is primarily the procedural overhead required to demonstrate compliance within those tiers, particularly for smaller organizations with fewer dedicated compliance resources.

The competitive pressure behind the deal

The simplification reflects explicit competitive anxiety within European policymaking. The United States under the current administration has pulled back from AI safety-focused federal regulation. China is deploying AI aggressively across manufacturing, transportation, and public services with minimal transparency requirements. European policymakers face a difficult calibration: a rigorous AI framework that discourages European AI development is worse than a moderate framework that sustains a European AI ecosystem capable of competing.

The political logic is harder than it sounds. Companies building in Europe face compliance costs that companies building in the U.S. or China don't. If those costs are high enough to push founders to incorporate in Delaware or Singapore instead of Berlin or Paris, the AI Act achieves oversight without generating a European AI industry to oversee.

Critics of the simplification argue the opposite: that the compliance requirements being eased are precisely the ones designed to catch harmful AI before it reaches people. Loosening them in response to industry lobbying repeats patterns seen in financial regulation before the 2008 crisis. Both positions are reasonable, which is why this is genuinely hard policy.

What hasn't changed

The August 2026 compliance deadline for high-risk AI systems has not moved. Companies deploying AI in covered categories — credit scoring, hiring tools, biometric identification, critical infrastructure management, educational assessment — still face substantive obligations including:

• Technical documentation of training data, model architecture, and performance metrics
• Human oversight mechanisms that allow meaningful review of high-stakes AI decisions
• Post-market monitoring requirements to detect performance degradation or bias drift
• Transparency obligations toward affected individuals

For multinational companies, the EU AI Act remains the de facto global compliance standard: a product meeting EU requirements typically exceeds what any other jurisdiction requires. That asymmetry has not changed. Building to EU standards is still the most efficient path to global compliance.

What it means for U.S. and global companies

American and Asian technology companies operating in Europe are watching the simplification closely, but the practical impact on large players is limited. The companies most relieved by reduced procedural burden are European SMEs and startups — the companies the EU actually needs to build a competitive domestic AI industry.

For large enterprises — whether European, American, or Asian — high-risk AI compliance was already a dedicated function with dedicated headcount. Streamlining conformity assessment procedures reduces cost at the margin but doesn't change the strategic calculus.

What to watch

The final simplified text will determine whether the changes are meaningful reductions in burden or primarily clarifications that don't move the needle operationally. The indicator group is European AI startups: if the simplification succeeds, European AI venture formation should hold competitive with the UK, Israel, and Canada rather than continuing to migrate toward U.S. jurisdictions. Watch Series A and Series B funding data for EU-incorporated AI companies through the end of 2026.