If you have been tracking AI regulation this year, you could be forgiven for thinking the ground has shifted under your feet twice over. In the European Union, the toughest obligations for high-risk AI systems have been pushed back by well over a year. In the United States, the federal government is trying to override state AI laws while those laws keep changing underneath it. For a business leader deciding how much to invest in AI governance right now, the temptation is to read both stories as "relax, there's more time." That reading is only half right, and acting on it fully could leave you exposed.
This article sets out what actually changed, what still applies today, and a practical governance baseline that holds regardless of which regulator blinks next.
What changed in the EU AI Act
The EU AI Act has been rolling out in phases since it entered into force in August 2024. Prohibited practices and AI literacy duties became enforceable on February 2, 2025, and GPAI provider obligations followed on August 2, 2025. The next big milestone was supposed to be high-risk AI system obligations under Annex III, originally due on 2 August 2026.
That changed in mid-2026. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026, six days before the original deadline. The deferral pushes compliance for standalone high-risk AI systems under Annex III from August 2026 to December 2, 2027, and for AI embedded in products already covered by EU product-safety law to August 2, 2028.
Crucially, the delay is selective. Compliance and governance teams should confirm that the unchanged August 2, 2026, Article 50 transparency deadline and the December 2, 2026, watermarking and Article 5 prohibition dates remain active on separate tracks. In other words, disclosure obligations for chatbots, deepfakes and AI-generated content were not delayed; only the heavier conformity-assessment regime for systems used in areas such as recruitment, credit scoring and education moved. If your generative AI use touches hiring, lending, insurance underwriting or similar high-stakes decisions, the extra runway is real, but it is runway for building the programme, not a reason to shelve it.
Why the US picture is messier, not calmer
In the United States there is no federal AI law, but there has been a concerted push to limit state ones. On December 11, 2025, President Trump signed the Ensuring a National Policy Framework for Artificial Intelligence executive order, a federal action designed to curtail the authority of individual states to regulate AI. The order directed the Attorney General to stand up an AI Litigation Task Force within 30 days to challenge state AI laws on dormant-commerce-clause, preemption and other constitutional grounds. A second order followed in June 2026 focused on cybersecurity and frontier-model access rather than state law.
The key point for any business operating across US states: an executive order cannot repeal a statute. Law firms tracking the litigation, including Paul Hastings and Latham & Watkins, are telling clients to keep complying with state AI laws as written, because an executive order cannot overturn a statute. No court has issued a final ruling invalidating any state AI law under the theories the executive order advances.
Colorado illustrates how fluid this is. The state's original AI Act was delayed twice, then effectively rewritten. On May 14, 2026, Colorado's governor signed SB 189, which revises the original AI law and delays the effective date from June 30, 2026, to January 1, 2027, while scaling back its requirements. The new version is narrower: it adopts a narrower approach focused on disclosures and transparency around automated decision-making technology, while preserving individual rights to access and correct data and to obtain meaningful human review of adverse automated decisions. On top of that, in April 2026 a major AI developer brought a federal constitutional challenge, the Department of Justice intervened in support, and a federal court entered a stay of enforcement on April 27, 2026. Several other states (California, Texas, Illinois among them) have their own AI-specific statutes with different triggers, so a single national posture is not available yet.
Why delays don't equal lower risk
Regulatory timelines are only one source of AI risk. Three others do not move with the EU or US calendar:
- Contractual exposure. Customers, insurers and enterprise procurement teams are increasingly writing AI-specific clauses into contracts (audit rights, liability caps, disclosure of training data sources) regardless of what the law requires.
- Reputational exposure. A biased hiring tool or a hallucinated customer communication causes damage long before a regulator investigates.
- Operational exposure. Ungoverned generative AI use, such as staff pasting client data into public chatbots, is a data protection and confidentiality problem under existing privacy law, not a future AI law.
Treating the EU deferral or the US preemption fight as a reason to pause is a mistake because none of these pressures have been deferred.
A governance baseline that survives the uncertainty
Regardless of which deadline eventually sticks, the following steps hold up:
- Inventory every generative AI use case, including shadow IT, and classify each by decision impact (does it touch hiring, credit, healthcare, or other consequential decisions?).
- Map each use case to the rules that already bite today: GPAI transparency if you deploy general-purpose models in the EU, Article 50 disclosure duties, and whichever US state laws apply to your operating footprint.
- Keep human review in the loop for any output that affects an individual's access to employment, credit, housing, healthcare or government benefits, since this remains a common thread across both Colorado's narrowed law and the EU's Annex III framework.
- Document data provenance and model choices now, even for use cases where the compliance deadline has moved. Conformity documentation is far cheaper to build as you go than to reconstruct retroactively.
- Revisit vendor and customer contracts for AI-specific warranties, audit rights and liability terms, since these are changing faster than statute.
- Assign clear ownership of the governance programme to a named individual or small committee, not a diffuse "everyone's responsibility" arrangement, so that when rules do firm up, there is a single point of accountability.
Conclusion
The EU AI Act's high-risk deadline has moved to December 2027 for most standalone systems, and the US federal government's attempt to override state AI laws has not yet succeeded in court or in Congress. Neither development removes the underlying commercial, contractual and reputational risks of deploying generative AI without governance. Businesses that use the extra time to build inventories, documentation and human oversight will be ready whenever the rules land; those that wait for certainty will be rebuilding from scratch under pressure. If you need help turning this into a concrete, workable programme, Wynstra can assist with designing governance and risk controls alongside the AI systems you are building or running.
Sources
- EU AI Act High-Risk Deadline Pushed to December 2027 - Cloud Security Alliance
- EU AI Act's High-Risk Deadline: Deferred, Not Cancelled - Cloud Security Alliance
- Colorado AI Act Amended and Effective Date Delayed - Hunton Andrews Kurth
- Colorado Repeals and Replaces Its Landmark AI Statute - Eckert Seamans
- New Executive Order Signals Evolving Federal Approach to AI - Lathrop GPM
- Trump AI Executive Order and State Law Preemption 2026 - SecurePrivacy
- President Trump Signs Executive Order Preempting State AI Laws and Centralizing Federal Oversight - Seyfarth Shaw