Prado AI / Executive Brief
AI GovernanceSeptember 21, 20267 min read

EU AI Act Update: High-Risk AI Deadlines Moved, But the Work Should Not Wait

The EU extended the deadline for high-risk AI rules, but this is not a reason to pause governance work. It is a chance to build the inventory, oversight, documentation, and vendor controls that will matter when enforcement arrives.

AI governance controls and regulatory timeline review

Most regulatory updates create one of two reactions inside a business. Either people panic because a deadline feels too close, or they relax because a deadline moved farther away.

The latest EU AI Act update deserves neither reaction.

In July 2026, the European Union’s AI Omnibus entered into force and changed the implementation timeline for high-risk AI systems. Under the updated timetable, rules for high-risk AI systems listed in Annex III apply from December 2, 2027. Rules for high-risk AI embedded in regulated products covered by Annex I apply from August 2, 2028. [1] [2]

That is more time. But for any company using AI in employment, credit, education, essential services, biometrics, critical infrastructure, or other sensitive areas, the practical message is not “wait.” The better message is “use the extra time properly.”

What Actually Changed

The EU AI Act already applies in phases. Some obligations are already in effect. Prohibited AI practices and AI literacy provisions began applying in February 2025. General-purpose AI governance obligations began applying in August 2025. Transparency rules, including disclosure obligations for certain AI systems and AI-generated content, started applying in August 2026. [3]

The AI Omnibus mainly adjusted the timing for high-risk AI systems. The revised timetable separates two categories.

For high-risk AI systems in Annex III, such as systems used in employment, education, access to essential services, law enforcement, migration, and certain biometric use cases, the rules apply from December 2, 2027. For high-risk AI embedded in regulated products, such as certain machinery, toys, lifts, and other product categories covered by Annex I, the rules apply from August 2, 2028. [4]

The European Commission describes the Omnibus as a simplification measure intended to give companies more legal clarity, expand sandbox access, reduce administrative burden for smaller businesses, and better align AI Act requirements with sectoral product rules. [1]

That is important. The deadline changed because the implementation environment needed more time, not because the underlying risk disappeared.

What Did Not Change

The basic direction of travel did not change. The EU is still moving toward formal governance requirements for higher-risk AI uses. The law still expects organizations to understand where AI is used, how risk is controlled, what data is involved, how human oversight works, and how incidents or errors will be detected.

The Commission’s AI Act materials describe high-risk obligations as including risk assessment and mitigation, high-quality datasets, activity logging, technical documentation, clear information for deployers, human oversight, robustness, cybersecurity, and accuracy. [5]

Those are not items a company can solve in the final month before a deadline. They require operational decisions. Who owns the system? What business process does it affect? What evidence is kept? What happens when the model is wrong? Which vendor controls are contractual, and which are only assumed?

The Omnibus gives businesses more time to answer those questions. It does not remove the questions.

Why This Matters for Small and Mid-Size Businesses

Many small and mid-size businesses assume the EU AI Act is mostly a large-enterprise issue. That assumption can be dangerous.

A local business may not be building an AI model from scratch, but it may still deploy AI through a vendor platform. A recruiting system may use AI to rank applicants. A lending workflow may use AI to score risk. An insurance workflow may use AI to classify claims or documents. A customer-service operation may use AI agents to handle requests. A productivity suite may introduce AI features into everyday decisions without a formal implementation project.

The compliance question is not “Did we build the AI ourselves?” The better question is “Are we using AI in a way that affects people, access, decisions, or rights?”

If the answer is yes, leadership should not wait until late 2027 to start documenting the environment. By then, the business may discover that AI is already spread across departments, vendor contracts, employee workflows, and customer-facing systems.

The Real Risk Is Not the Deadline. It Is the Inventory Gap.

The most common governance weakness is simple: nobody has a complete list of where AI is being used.

That gap is easy to understand. AI is no longer a single project owned by IT. It appears inside customer service tools, office productivity suites, marketing platforms, human resources systems, analytics dashboards, cybersecurity products, and document-processing workflows. Employees also use external AI tools informally because they make everyday work faster.

Without an AI inventory, a business cannot classify risk. Without risk classification, it cannot decide which systems need stronger oversight. Without oversight, the business is relying on individual judgment, vendor defaults, and luck.

A basic AI inventory does not need to be complicated. It should identify the tool, vendor, business owner, purpose, data used, users affected, decision impact, human review point, and whether the system touches a sensitive area such as employment, credit, insurance, healthcare, education, or access to services.

That inventory becomes the foundation for every later compliance step.

High-Risk Does Not Mean Every AI Tool

It is also important not to overreact. Not every AI tool is high-risk under the EU AI Act. Many AI uses are low-risk or limited-risk. Drafting an internal email, summarizing a meeting, organizing notes, or helping an employee prepare a first draft will usually not create the same regulatory concern as using AI to screen job candidates or determine access to a loan.

That distinction matters because good governance should be proportionate. A company that treats every AI use as equally severe will create unnecessary bureaucracy. A company that treats every AI use as harmless will miss the systems that deserve control.

The practical goal is to separate ordinary productivity use from AI that materially affects people, regulated processes, or customer outcomes.

What Businesses Should Do Now

The next step is not to write a long AI policy that nobody reads. The next step is to build a working control base.

Start with an AI inventory. List the systems already in use, including vendor tools with AI features built in. Then classify each system by business impact. Does it support internal productivity, influence a human decision, or automate a decision that affects a customer, employee, applicant, or third party?

After that, identify the systems that need human oversight. The oversight point should be specific. “A manager reviews it” is weaker than “the operations manager reviews all AI-flagged exceptions before a customer receives a final decision.”

Next, review vendor contracts and documentation. Many businesses rely on vendor claims without checking whether the vendor provides audit logs, data-use boundaries, model documentation, incident support, or a clear explanation of where customer data goes.

Finally, keep evidence. If an AI system supports an important business process, the company should be able to show who owns it, what it does, what data it uses, what review occurs, and what happens when something goes wrong.

That is governance in practical terms. It is not a binder. It is operational discipline.

The Bottom Line

The EU AI Omnibus moved high-risk AI deadlines to December 2, 2027 for Annex III systems and August 2, 2028 for high-risk AI embedded in regulated products. That gives businesses more time, but it should not create complacency.

The companies that benefit from the delay will be the ones that use it to understand their AI footprint, identify high-impact uses, assign ownership, and build evidence before the deadline arrives.

For small and mid-size businesses, the right question is not “Are we ready for a regulator?” It is simpler and more useful: “Do we know where AI is affecting our business, our customers, and our decisions?”

If the answer is unclear, that is the place to start.

This article provides general business information, not legal advice. Organizations should obtain advice that reflects their specific AI uses, markets, and regulatory obligations.

Sources

[1] European Commission, “AI Omnibus enters into force”
[2] EUR-Lex, “Regulation (EU) 2026/1744”
[3] European Commission AI Act Service Desk, “Timeline for the Implementation of the EU AI Act”
[4] European Commission, “The enforcement framework of the AI Act”
[5] European Commission, “AI Act regulatory framework”

Have a question about this article?

Chat with our team directly on WhatsApp.

Chat on WhatsApp
PP

Pedro Prado

Founder, Prado AI Solutions. 30+ years in operational risk, audit, and controls. Published author on AI. Trilingual: English, Spanish, Portuguese.

Prado AI / Next Step

Ready to put this into practice?

Get a written AI readiness assessment identifying your top opportunities and governance gaps, at no cost.

Get My Free AI Readiness Audit
Ask us about
our services