On 16 June 2026, the European Parliament voted 423 to 57 (with 174 abstentions) to approve the first amendments to the EU AI Act since it entered into force in August 2024. The vote adopted the “Digital Omnibus on AI” package: it pushes back the compliance deadline for high-risk AI systems to 2 December 2027 (or 2 August 2028 for AI embedded in regulated products), extends the AI-content watermarking deadline to 2 December 2026, and creates a new EU-wide ban on “nudifier” apps and AI-generated child sexual abuse material, also taking effect 2 December 2026.
The European Parliament AI Act vote 2026 is a single, targeted amendment procedure — not a rewrite of the regulation. The AI Act’s risk-based architecture, its prohibited-practice list, and its general-purpose AI (GPAI) obligations are unchanged and still apply from 2 August 2026. What moved is the compliance runway for the highest-obligation tier, plus one significant new prohibition.
The Digital Omnibus on AI is the legislative vehicle the European Commission proposed on 19 November 2025 as part of its seventh simplification “omnibus” package, following a provisional political agreement between Parliament, Council, and Commission negotiators on 7 May 2026. The Council of the European Union has stated that, because obligations on high-risk AI systems were due to enter into force on 2 August 2026, the co-legislators treated the file with priority. The Council must still give its own formal sign-off — expected before that 2 August 2026 date — before the amendments take legal effect.
What exactly did the Parliament vote pass?
MEPs approved targeted amendments to Regulation (EU) 2024/1689 (the AI Act), submitted under procedure file 2025/0359(COD). The result — 423 in favour, 57 against, 174 abstaining — is a decisive majority, but the near-40% abstention rate signals real political division over loosening any part of a regulation adopted only two years earlier.
Co-rapporteur Arba Kokalari (EPP, Sweden) framed the vote as “pressing the pause button on the AI Act” to cut red tape for companies, while co-rapporteur Michael McNamara (Renew, Ireland) described it as “establishing legal certainty by extending certain timelines while preserving the AI Act’s architecture.” Both statements matter for compliance planning: the vote is explicitly a timeline and scope adjustment, not a reopening of the Act’s risk classifications or its prohibited-practices list.
New deadlines: what got delayed, and until when
The amendment postpones application of high-risk AI obligations to give standard-setting bodies and market-surveillance authorities time to finish the technical standards the Act depends on. Three separate dates now apply, replacing the single 2 August 2026 cut-off in the original text.
| Obligation | Original deadline | New deadline (as passed 16 June 2026) |
|---|---|---|
| Stand-alone high-risk AI systems (Annex III — e.g. hiring, education, law enforcement) | 2 August 2026 | 2 December 2027 |
| High-risk AI embedded as a safety component in regulated products (e.g. medical devices, machinery) | 2 August 2026 | 2 August 2028 |
| AI-generated content watermarking/labelling (systems on the market before 2 August 2026) | 2 August 2026 | 2 December 2026 |
| Ban on nudifier and CSAM-generating AI systems | Not previously in the Act | 2 December 2026 |
Crucially, what did not move: the AI Act’s prohibited-practices provisions (e.g. social scoring, manipulative AI) and its general-purpose AI transparency obligations remain scheduled for 2 August 2026, and AI-literacy requirements for staff working with AI systems are already in force. Institutions should not read the high-risk delay as a delay to the whole Act.
The nudifier app ban: what is now illegal
The amendment inserts an outright ban on AI systems that generate child sexual abuse material or produce images, video, or audio depicting an identifiable person’s intimate parts or sexual activity without consent. Providers cannot place such systems on the EU market unless they build in technical safeguards against that use; deployers who use existing tools for this purpose are equally liable. Compliance is required by 2 December 2026.
McNamara said the ban targets tools that “impact real people, overwhelmingly women, with the purpose of humiliating, degrading and objectifying them,” and confirmed Parliament fought to have it “enter into force before the end of this year.” This is the amendment’s only genuinely new prohibition — everything else in the package either delays or clarifies existing obligations.
Other changes: machinery, bias testing, SME exemptions
Beyond deadlines and the nudifier ban, the adopted text makes four further adjustments:
- Removes duplicate compliance requirements for AI-enabled machinery products, so manufacturers satisfy sectoral safety law rather than both that law and the AI Act in parallel.
- Narrows the definition of “safety component,” so AI features that only assist users or optimise performance no longer automatically trigger high-risk obligations unless their failure poses a genuine health or safety risk.
- Permits processing of sensitive personal data, under safeguards, specifically to detect and correct bias in both high-risk and non-high-risk AI systems.
- Extends existing SME compliance exemptions to “small mid-cap enterprises” (SMCs), and centralises enforcement of certain general-purpose AI systems within the EU AI Office.
What this means for research institutions and publishers
Most Digital Omnibus coverage has focused on industry and machinery. Two effects are more directly relevant to research administrators, universities, and scholarly publishers, and are largely absent from legal-sector explainers of this vote.
First, the sixteen-month extension for standalone high-risk AI systems covers tools universities have begun deploying for candidate screening, admissions triage, and staff performance evaluation — Annex III “employment and worker management” use cases. Institutions now have until 2 December 2027, not 2 August 2026, to reach conformity, but the underlying obligation, and the requirement that staff have adequate AI literacy, has not been removed.
Second, the extended watermarking deadline (2 December 2026 for tools already on the market) affects AI-detection and provenance workflows journals use to flag AI-generated text and images in submissions. Publishers should not assume machine-readable AI-content labels are universally present in EU-origin content before that date, since pre-existing tools get an extra four months’ grace.
Neither effect changes CASRAI’s role: CASRAI originated the CRediT contributor role taxonomy in 2014, and the standard is now stewarded by NISO as ANSI/NISO Z39.104-2022. The Digital Omnibus does not touch authorship or contributorship standards directly, but institutions mapping AI-assistance disclosure into research administration workflows should track the December 2026 and December 2027 dates alongside their own policy review cycles.
Answer-first Q&A
What did the European Parliament vote on for the AI Act in June 2026?
On 16 June 2026, MEPs voted 423 to 57 to approve the Digital Omnibus on AI, a targeted amendment package that delays several AI Act compliance deadlines and adds a new ban on AI “nudifier” apps and CSAM generation, without altering the Act’s risk-based structure.
When do the EU AI Act’s high-risk rules now apply?
Under the amended text, stand-alone high-risk AI systems must comply from 2 December 2027, and high-risk AI embedded in regulated products such as medical devices or machinery must comply from 2 August 2028 — both later than the original 2 August 2026 date.
What is banned under the EU’s nudifier app ban?
The ban covers AI systems that generate non-consensual intimate imagery of an identifiable person or child sexual abuse material. It applies to developers and deployers alike and takes effect 2 December 2026, six months after the Parliament vote.
Does the Digital Omnibus change the AI Act’s risk-based approach?
No. Parliament’s own summary states the amendments aim to support compliance “while maintaining the law’s main provisions and risk-based approach.” Prohibited practices and general-purpose AI transparency duties are unchanged and remain due 2 August 2026.
What happens next
The text adopted on 16 June 2026 still requires formal adoption by the Council of the European Union before it enters into force — a procedural step Parliament’s own press release flags under “next steps.” That approval is widely expected before 2 August 2026, the date on which most of the AI Act’s remaining original provisions still take effect regardless of this amendment. Institutions should treat the new 2027/2028 dates as the operative planning horizon for high-risk systems, while continuing to meet every obligation the vote left untouched.