Recommended Articles
Legal & Policy

EU AI Act High-Risk Delay: What Founders Must Still Build Before December 2027

By Angeled · November 1, 2026 · Legal & Policy
EU AI Act High-Risk Delay: What Founders Must Still Build Before December 2027
EU AI Act High-Risk Delay: What Founders Must Still Build Before December 2027

Europe has given artificial intelligence companies more time on the hardest part of its landmark AI law, but the delay is partial, not a pause. The high-risk obligations in the EU AI Act have been pushed back to at least December 2027, just as the European Commission has referenced the first harmonised standard under the law. For founders building AI products for European customers, the message is subtle: the deadline moved, but the work did not disappear, and several duties are already enforceable.

The change comes from the Digital Omnibus on AI, formally Regulation (EU) 2026/1744. Under the amended timetable, obligations for stand-alone high-risk systems listed in Annex III now apply from 2 December 2027, rather than 2 August 2026. High-risk AI embedded in products covered by EU product-safety law under Annex I applies from 2 August 2028. Annex III covers sensitive uses such as recruitment, credit scoring, education, law enforcement, and critical infrastructure. Those are exactly the markets where a compliance failure can end a customer relationship, not merely attract a fine.

Why the dates moved

The stated reason is practical rather than political. Harmonised standards, common specifications, guidance, and national enforcement bodies were not ready on the original timetable. The Commission's standardisation request to CEN and CENELEC dates to May 2023, with an April 2025 deadline that was missed, and the remaining standards were pushed toward the end of 2026. As of late July 2026, no harmonised standard for the AI Act had been cited in the Official Journal. That changed in October 2026, when the Commission published the reference to the first such standard: a quality management system standard from CEN and CENELEC's JTC 21.

A harmonised standard matters because it gives providers a presumption of conformity. In plain language, a company that follows the standard is presumed to meet the corresponding legal requirements, unless evidence shows otherwise. The quality management standard covers how AI systems are designed, documented, and monitored. It is therefore a useful backbone for a startup's engineering process, even before the high-risk duties formally apply. It does not, however, turn compliance into a box-ticking exercise. Risk management, human oversight, accuracy, robustness, and post-market monitoring remain substantive obligations.

What was not delayed

The delay covers high-risk obligations. It does not erase the rest of the Act. Prohibited practices and the AI literacy duty have applied since 2 February 2025. General-purpose AI model obligations have applied since 2 August 2025. Transparency obligations under Article 50 became generally applicable on 2 August 2026. Those transparency rules matter for a much wider group of companies than the high-risk chapter does, because they apply regardless of whether a system is classified as high-risk.

Article 50 covers, among other things, how certain AI-generated or manipulated content is identified. If a product talks to users, generates images, audio, video, or text, or scores emotions or biometrics, founders should assume the transparency duties deserve a legal review now. Violations can carry fines of up to 15 million euros or 3 percent of worldwide annual turnover under Article 99(4)(g), according to reporting on the provision. The Omnibus left the Article 50 date alone, with only a four-month transition on the machine-readable marking duty in Article 50(2) for generative systems already on the market before 2 August 2026, ending on 2 December 2026.

For a startup, that bargain has a clear implication. The law is not asking for less safety; it is asking for safety demonstrated through the company's own systems, documentation, and monitoring, rather than through a single prescribed form. That favours teams that build evidence as they build product: versioned model cards, evaluation results, data provenance notes, incident logs, and human-oversight records. Teams that treat compliance as a document to be written at the end will find the December 2027 date arrives faster than expected.

Why it matters for founders and investors

Investors diligence AI startups differently in Europe now. A credible answer to "what is your AI Act plan?" should distinguish three layers: what is already in force, what is delayed, and what depends on standards still being published. A pitch that says "the AI Act was delayed, so we have nothing to do" will read as naive. A pitch that shows transparency labelling, a quality management process aligned to the first harmonised standard, and a dated roadmap to Annex III readiness will read as fundable.

Founders should also price the delay honestly. More time lowers near-term legal spend, but it does not lower the eventual engineering cost. Hiring, procurement, and enterprise sales cycles in regulated sectors will still ask for conformity evidence long before December 2027. Early customers in recruitment, lending, education, and healthcare will want to see risk assessments and monitoring plans in the sales process, not after signature.

Angeled.com connects founders with investors.

Source: tecniforge.com
Source: techvify.com
Source: goodwinlaw.com

Read Next