The 2027 and 2028 safeguards

The practical judgement

The 2027 and 2028 rules mainly strengthen the systems surrounding high-risk AI. They impose risk controls, information, oversight and assessment duties on organisations. They do not guarantee that an affected person can demand a different outcome.

Regulation (EU) 2026/1744 entered into force on 27 July 2026 and changed earlier dates. As of 12 September 2026, the enacted dates are 2 December 2027 for Annex III high-risk systems and 2 August 2028 for high-risk AI embedded in Annex I regulated products.¹

The following figure adds the deferred safeguards to the 2026 field and shows who bears them.

<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 590 440" width="590" height="440" font-family="system-ui, sans-serif" font-size="14">
<title>Deferred high-risk AI safeguards added in 2027 and 2028</title>
<rect x="20" y="18" width="550" height="62" rx="8" fill="none" stroke="currentColor" stroke-width="2"/>
<text x="295" y="44" text-anchor="middle" fill="currentColor" font-size="16">2026: prohibitions, transparency, model duties, complaints</text>
<text x="295" y="67" text-anchor="middle" fill="currentColor" font-size="16">Explanations: practical reach unresolved until 2 December 2027</text>
<line x1="295" y1="80" x2="295" y2="116" stroke="currentColor" stroke-width="2"/>
<polygon points="295,116 288,104 302,104" fill="currentColor"/>
<rect x="20" y="118" width="265" height="224" rx="8" fill="none" stroke="currentColor" stroke-width="2"/>
<text x="152" y="146" text-anchor="middle" fill="currentColor" font-size="16">Annex III from Dec 2027</text>
<text x="42" y="180" fill="currentColor">Provider risk controls</text>
<text x="42" y="212" fill="currentColor">Designed human oversight</text>
<text x="42" y="244" fill="currentColor">Trained deployer oversight</text>
<text x="42" y="268" fill="currentColor">Notice for systems making or</text>
<text x="42" y="290" fill="currentColor">assisting decisions about people</text>
<text x="42" y="318" fill="currentColor">Selected impact assessments</text>
<rect x="305" y="118" width="265" height="224" rx="8" fill="none" stroke="currentColor" stroke-width="2"/>
<text x="437" y="146" text-anchor="middle" fill="currentColor" font-size="16">Annex I from Aug 2028</text>
<text x="327" y="180" fill="currentColor">Provider risk controls</text>
<text x="327" y="212" fill="currentColor">Designed human oversight</text>
<text x="327" y="244" fill="currentColor">Trained deployer oversight</text>
<text x="327" y="276" fill="currentColor">Product-embedded systems</text>
<text x="327" y="308" fill="currentColor">No Article 86 expansion</text>
<line x1="152" y1="342" x2="152" y2="378" stroke="currentColor" stroke-width="2"/>
<line x1="437" y1="342" x2="437" y2="378" stroke="currentColor" stroke-width="2"/>
<line x1="152" y1="378" x2="437" y2="378" stroke="currentColor" stroke-width="2"/>
<line x1="295" y1="378" x2="295" y2="400" stroke="currentColor" stroke-width="2"/>
<polygon points="295,400 288,388 302,388" fill="currentColor"/>
<rect x="90" y="400" width="410" height="34" rx="8" fill="none" stroke="currentColor" stroke-width="2"/>
<text x="295" y="423" text-anchor="middle" fill="currentColor">Operator safeguards, not guaranteed personal reversal</text>
</svg>

Figure: Deferred dates add provider and deployer safeguards, with different effects for Annex III and Annex I systems.

Under Article 16, providers must ensure that high-risk systems meet Articles 8 to 15, covering risk management, data governance, logging, documentation, instructions, oversight, accuracy, robustness and cybersecurity. National market-surveillance authorities or the AI Office within its remit enforce that provider duty from 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems.²

Articles 14 and 26 require providers to design effective oversight and deployers to appoint people with suitable competence, training, authority and support. The same authorities enforce those respective duties from 2 December 2027 for Annex III and 2 August 2028 for Annex I.²

For Annex III systems that make or assist decisions about people, Article 26 requires the deployer to tell affected people that the system is being used. National market-surveillance authorities or the AI Office within its remit enforce that deployer duty from 2 December 2027.² Article 27 also requires public bodies, public-service providers and deployers of specified creditworthiness and life or health insurance systems to assess fundamental-rights effects before deployment. Those authorities enforce the covered deployer’s assessment duty from the same date.²

Article 86 applied from 2 August 2026, but its practical reach remains unresolved until 2 December 2027 because Annex III classification was deferred.⁴ The deployer bears the explanation duty, enforced by market-surveillance authorities or the AI Office within its remit. It applies only to adverse decisions with legal or similarly significant effects and excludes specified critical-infrastructure systems.²

You need not wait to use the GDPR where personal data support a decision. Its access, rectification, objection, automated-decision, complaint and court routes already apply and can continue alongside applicable AI Act mechanisms.³

Boundary. The 2028 Annex I rules do not expand Article 86, which is confined to Annex III. Article 111 also limits the new regime for unchanged non-public systems already in service: providers and deployers comply only after a significant design change, with the relevant market-surveillance authority or AI Office enforcing from the applicable 2027 or 2028 date. Pre-existing systems intended for public-authority use have a separate 2 August 2030 compliance date under Article 111, enforced by those authorities.²