The European Union has formally pushed back the most demanding parts of its AI Act, but it has not paused the whole law. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on July 24, 2026, and took effect on July 27. It moves the deadline for standalone high-risk AI systems to December 2, 2027. Companies still face a hard date on December 2, 2026, when new bans and labeling requirements begin to bite.
What the Omnibus Changed
The original Act set August 2, 2026 as the date when high-risk obligations would apply to standalone systems. Under the Omnibus, those obligations now start on December 2, 2027. Those systems include AI used in employment, credit decisions, education and biometrics. AI embedded in regulated products, such as medical devices, machinery, toys and vehicles, moves from August 2, 2027 to August 2, 2028. Systems used by public authorities have until August 2, 2030.
According to ActuIA's breakdown of the regulation, the amendment changes the timing of the high-risk regime without repealing it. The rules remain in the Act. What changes is how long companies have to prepare. Softtechhub calls the delay "a concession to reality rather than a reversal."
What Did Not Move
The delay is narrower than the headline may suggest. The prohibitions in Article 5 of the Act and the AI literacy obligation, which took effect in February 2025, remain in place. The obligations for general-purpose AI models, which started in August 2025, were also left alone. Transparency duties under Article 50, including the requirement to tell people they are interacting with an AI system, still apply from August 2, 2026, according to Softtechhub's analysis of the changes.
That makes the regime more layered than a single date. A company running a chatbot in Europe may already have transparency duties, while the high-risk rules for hiring or credit tools are now more than a year away.
The December 2 Deadline
The date companies should circle is December 2, 2026. Two things happen then. First, providers of generative AI systems that were already on the market before August 2, 2026 must comply with the machine-readable marking requirement in Article 50(2). That requirement means synthetic outputs need to carry a format that software can detect. The grace period ends on that date for older systems.
Second, two new prohibitions take effect. Under the amended Article 5, the Act now bans systems that generate or manipulate realistic images, videos or audio of an identifiable person's intimate parts or sexually explicit activity without that person's explicit consent. A separate ban covers systems that produce child sexual abuse material as defined in existing EU law. For providers, the second ban applies where that output is the intended purpose or a reasonably foreseeable result without adequate safeguards. ActuIA notes that recital 11 of the regulation cites so-called nudification apps as a target.
These bans are the part of the Omnibus that is not a delay. They are new restrictions that take effect in a few weeks, and they apply to developers and deployers alike.
Enforcement Is Still Uneven
Delayed deadlines do not mean the existing rules are dormant. Softtechhub reports that enforcement of the tiers already in force has been light in practice, but that the legal exposure is real. The AI Office has gained expanded powers, including the ability to open investigations, carry out on-site inspections, accept binding commitments and impose fines. Penalties for the most serious prohibition violations can reach 35 million euros or 7% of global turnover.
That combination creates a practical question for companies. Many are using the extra time to build compliance programs, but the most serious penalties attach to the bans that are already in force or are about to start. Waiting for the high-risk deadline is a plan only for systems that do not touch those bans.
Why It Matters
The Omnibus is a signal that Brussels is willing to adjust timelines when standards and guidance lag behind the law. It is also a sign that the EU will still set hard lines on the most harmful uses of generative AI. For companies selling AI products in Europe, the lesson is to separate the timelines into two groups: the bans and transparency duties that apply now or within weeks, and the high-risk rules that have more room.
For the wider AI policy debate, the Omnibus offers a case study. Lawmakers can slow the compliance clock while keeping the core obligations, and they can add targeted prohibitions in response to specific harms, such as non-consensual intimate imagery. Other jurisdictions that are drafting their own rules are likely to watch how the EU handles that balance over the next year.
