One Vote Left: Why August 2 Still Matters More Than the Omnibus

🏛️ The Legislative Marathon Is Nearly Over

On 16 June 2026, the European Parliament approved the Digital Omnibus on AI with 423 votes in favour, 57 against, and 174 abstentions, as confirmed in the European Parliament’s official announcement. The EU Council is expected to formally adopt the text on 29 June — the last procedural step before publication in the Official Journal, followed by entry into force three days later.

That Council vote is not merely procedural. The deferred high-risk dates apply only if the Omnibus enters into force before 2 August 2026, when current high-risk rules would otherwise take effect. Until then, businesses technically remain subject to the original statutory timetable, even though formal adoption before August is widely expected according to both the Council of the European Union and independent legal analyses.

📅 What the Final Deal Actually Locks In

The Omnibus restructures, not dismantles, the AI Act. The timetable below reflects the final political agreement described by the Council of the European Union and subsequent legal analysis from White & Case.

* Applies only to certain AI systems placed on the market before 2 August 2026.

August 2 remains an active compliance date. Human-facing AI disclosures, chatbot transparency requirements, and most Article 50 obligations remain on the original schedule; only a limited subset of machine-readable watermarking obligations receives a short extension.

⚖️ August 2 Is Live — What Activates

August 2, 2026 is the date the Commission gains penalty enforcement powers over GPAI providers, Article 50 transparency obligations activate for most AI systems, and competent authorities obtain their full enforcement toolkit for provisions applicable from August 2. Penalty tiers are unchanged: €35M/7% of global turnover for prohibited practices, €15M/3% for high-risk and transparency violations, and €7.5M/1.5% for misleading information to authorities.

This follows directly from what we covered in Europe’s AI Labelling Clock Is Ticking. The Council vote this week removes most remaining legislative uncertainty: compliance teams now have a single, fixed planning calendar. For organisations already navigating agentic deployments — where the governance gap is already wide — Article 50 deployer disclosure applies to any AI-facing customer or worker interface. August 2 is not a future problem.

🧭 The Classification Consultation: A Hidden Signal

The Commission has extended its high-risk classification consultation deadline to 23 July 2026, with final guidelines expected by the end of 2026 as part of the Commission’s AI Act implementation programme.

The extension matters more than it looks. The draft guidelines interpret the high-risk conformity assessment test broadly, capturing more AI systems than businesses may have assumed. Providers should exercise care in publicly available materials — instructions for use, technical documentation, and marketing — because those descriptions may become important evidence when authorities assess whether a system has been correctly classified as high-risk.

The analytic risk here is scope creep by documentation. Companies that write broad capability claims into product materials may inadvertently strengthen the case for high-risk classification. July 23 is the last formal window to push back before the Commission finalises its interpretive framework.

📉 The Diminishing-Support Pattern

One detail the headlines underplay: political consensus is narrowing. On 26 March 2026, Parliament adopted its negotiating position with 569 votes in favour; on 16 June, the agreed text recorded 423 in favour, 57 against, and 174 abstentions — a near-quadrupling of abstentions.

The voting pattern suggests noticeably weaker political consensus than earlier negotiations. Civil society has criticised the fast-track Omnibus process as weakening digital rights protections, while EU institutions maintain that the amendments simplify implementation without altering the AI Act’s core objectives.

This governance fragility has a clear parallel: GDPR’s Article 29 Working Party guidance repeatedly lagged legal deadlines, creating enforcement gaps that persisted for years. The AI Act’s classification guidance may face similar implementation challenges — final guidelines due end-2026, obligations live by December 2027. Eighteen months is enough time to act; it is not enough time to wait.

👀 What to Watch

The Council is expected to adopt the Omnibus on June 29, completing the legislative process before publication in the Official Journal and entry into force three days later, assuming the current timetable holds.

The real watch item is whether the Commission’s high-risk classification guidelines — finalised after the July 23 consultation — define the Article 6(3) exception narrowly or permissively. A narrow reading compresses the relief the Omnibus appears to offer. Compliance teams building their inventory now, before final guidance lands, are making the right call.

My read: The Digital Omnibus is no longer the main story. Once formally adopted, the debate shifts from when the AI Act applies to how organisations operationalise Article 50. The legislative uncertainty is largely over; the implementation challenge has arrived. Providers still have only weeks—not years—to ensure customer-facing disclosures, synthetic-content handling, governance processes, and enforcement readiness are in place before August 2. That is now the date boards, legal teams, and AI governance leaders should be planning against.