The EU AI Act Is Finally Here: What Changes In 2026 And What Can Wait Until 2027?

For what feels like years, the EU AI Act has been hanging over the technology industry. Everybody’s been talking about it, speculating about the specifics and trying to plan for the future.

Startups have worried about compliance costs, big AI companies have lobbied for delays and regulators have spent months publishing guidance and preparing enforcement structures to come into force.

And now, at last, one of the Act’s biggest milestones has finally arrived. As of 2 August 2026, the Act is finally in place and the legislation has come into force. But, despite some dramatic headlines, this still doesn’t mean every part of the AI Act applies immediately.

In fact, some of the rules that businesses have been most concerned about actually won’t apply until December 2027, with others delayed until as late as 2028. And for startups and smaller businesses especially, knowing which rules and deadlines will be implemented when is absolutely essential.

Indeed, as we’ve previously covered on TechRound, opinion on the AI Act remains divided. Some people see it as an important step towards safer and more trustworthy AI, while others worry that it could create extra burdens for European companies that are competing in a fast-moving global market.

And now, at least, the latest deadline has arrived, and the focus is shifting from speculation to implementation.

What Has Changed As of 2 August 2026?

According to the European Commission’s AI Act Service Desk, 2 August 2026 marks the point at which the majority of the Act’s remaining provisions begin to apply and enforcement formally starts for the rules that were already in force.

Most importantly for many businesses, the AI Act’s transparency requirements under Article 50 are now applicable. And for many, this is one of the most important parts.

These rules are designed to ensure that people know when they are interacting with AI systems and when content has been generated or manipulated by AI. The requirements cover areas such as AI-generated content, synthetic media and certain AI systems that interact directly with users.

The Commission also notes that measures that are designed to support innovation begin applying from this date, whilst national and EU-level authorities can now formally enforce applicable provisions relating to prohibited AI practices, transparency requirements, AI literacy obligations and general-purpose AI models.

So in short, August 2026 is the point at which the AI Act becomes a practical reality for many businesses. Basically, it goes from being a looming future compliance exercise to the reality of business.

What Was Already In Force?

One reason many people are finding the AI Act and all its relevant dates confusing is that it’s been rolled out gradually.

The first major implementation phase arrived on 2 February 2025, when the Act’s definitions, AI literacy provisions and prohibited AI practices became applicable.

The next milestone followed on the 2nd of August 2025, when governance requirements and obligations for providers of general-purpose AI models came into force. At this point, Member States were also required to establish governance structures and national authorities responsible for oversight.

What does all this mean? Basically, it means that many of the foundations of the legislation have actually already been in place for some time.

What’s Been Delayed Until 2027?

This is where many startups may find some reassurance. Following changes introduced through the AI Omnibus package, the timetable for high-risk AI systems has been pushed back. The rules for standalone high-risk AI systems that were covered under Annex III of the legislation will now apply from 2 December 2027 rather than August 2026, so there’s a little more time to prepare.

These include certain AI systems used in areas such as employment, education, access to essential services, critical infrastructure, migration and law enforcement. The delay was introduced because key standards, guidance and support tools weren’t yet fully available (wouldn’t have been available in time), which unsurprisingly created concerns about how businesses could realistically comply with the rules.

Ultimately, the revised timeline is intended to give both regulators and companies more time to prepare and ensure that everything’s done properly, responsibly and safely.

What About AI Regulatory Sandboxes?

Another important date for startups is 2 August 2027, almost exactly a year from now. Under the revised timetable, Member States are expected to have at least one AI regulatory sandbox operational by 2 August 2027.

These sandboxes are designed to allow companies to develop and test AI systems, but all within a supervised regulatory environment, helping businesses understand compliance requirements before bringing products to market. This is supposed to contribute to additional safety precautions.

For early-stage companies, they could become one of the most useful parts of the AI Act framework.

What Happens In 2028?

The final major implementation stage arrives on 2 August 2028: the requirements for high-risk AI systems embedded within regulated products become applicable.

These are the systems that are classified as high-risk because they form part of products covered by existing EU product safety legislation. That means things like certain medical devices, machinery and other regulated equipment. Basically, things that are considered high risk.

So What Should Startups Focus On Today?

Ultimately, most startups don’t need to prepare for every aspect of the AI Act immediately, and they don’t need to panic.

The August 2026 milestone is significant because transparency requirements are now in force and enforcement has formally begun for applicable provisions. But still, many of the high-risk obligations that generated the most concern have been pushed back until December 2027 or August 2028, so there’s still a fair bit of time to get ducks in a row.

For founders, the best approach would be to focus on what applies now while still keeping an eye on what’s coming next.

Now, the AI Act is no longer a future regulation: it’s here and it’s real. But for many startups, the most demanding compliance requirements are still more than a year away. That doesn’t mean they should sit back, relax and do nothing. But

That breathing room may prove valuable for a sector that has spent much of the last two years wondering exactly when the rules would finally arrive.