EU AI Act Fines Before High-Risk Obligations Apply in December 2027

December 2027 is not the start of exposure

Many compliance programmes treat the high-risk deadline as the point at which the EU AI Act becomes enforceable. That reading is incomplete. Regulation (EU) 2026/1744, the Digital Omnibus on AI of 8 July 2026, moved the application date of Chapter III Sections 1, 2 and 3 from the original 2 August 2026. High-risk systems under Article 6(2) and Annex III now fall under those requirements from 2 December 2027. High-risk systems under Article 6(1) and Annex I follow on 2 August 2028. The stated reasons are the delayed availability of standards, common specifications and guidance, and the delayed establishment of national competent authorities.

The postponement concerns those high-risk requirements. The Commission's own pages continue to list the earlier dates for other parts of the Act. Prohibited practices and AI literacy obligations have applied since 2 February 2025. Governance rules, penalty provisions and GPAI model obligations have applied since 2 August 2025. A regulated enterprise that uses AI today can therefore already be within reach of the penalty regime, depending on its role and the use in question.

Application dates that matter for enforcement exposure

The dates below separate what already applies from what is deferred. Status is stated as of November 2026 and could change if the legal texts are amended again.

• 2 February 2025: Prohibitions, definitions and AI literacy obligations apply.
• 2 August 2025: Governance rules, penalty provisions and GPAI provider obligations apply.
• 2 August 2026: The Commission enforces GPAI provider obligations, including through fines.
• 2 August 2027: GPAI models placed on the market before 2 August 2025 must comply.
• 2 December 2027: Chapter III Sections 1 to 3 apply to Annex III high-risk systems.
• 2 August 2028: Chapter III Sections 1 to 3 apply to Annex I high-risk systems.

Penalty tiers that can apply now

Article 99 sets administrative fines as a fixed amount or a percentage of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs, the lower of the two amounts applies in each category. The EUR-Lex summary also refers to proportionality adjustments for SMEs, including start-ups, and small mid-caps.

Penalties must be effective, proportionate and dissuasive. Member States must take the Commission's Article 96 guidelines into account when applying them. Penalties can also include warnings and non-monetary measures.

• Article 99(3): prohibited practices: Non-compliance with the Article 5 prohibitions: up to EUR 35,000,000 or, for an undertaking, up to 7% of worldwide annual turnover, whichever is higher.
• Article 99(4): other operator obligations: Non-compliance with listed provisions on operators or notified bodies other than Article 5: up to EUR 15,000,000 or up to 3% of worldwide annual turnover, whichever is higher. The Commission describes this tier as covering the other requirements and obligations of the Regulation.
• Article 99(5): information to the AI Office: Supplying incorrect, incomplete or misleading information to the AI Office in reply to a request is subject to fines.
• Corrective periods: If a provider does not bring an AI system into compliance within the period set by a market surveillance authority, it is subject to fines under Article 99.
• AI Office decisions: Fines may accompany AI Office decisions, applying Article 99(3) to (7) mutatis mutandis. Periodic penalty payments may not exceed 5% of average daily income or worldwide annual turnover per day.

Who enforces what

Enforcement is shared. Member States must lay down the rules on penalties and other enforcement measures for infringements by operators, so deployers and most other operators are mainly exposed through national penalty rules and the authorities each Member State designates. The Omnibus cites the delayed establishment of national competent authorities, so the position can differ between Member States. An enterprise operating across several should check which authority and which penalty rules apply in each.

The Commission and the AI Office have a separate role for GPAI model providers. The Commission states that it will enforce GPAI provider obligations, including through fines, from 2 August 2026. The amended Act allows the AI Office to impose fines and periodic penalty payments, and the Court of Justice has unlimited jurisdiction to review them. An AI Office infringement decision may cover any applicable provision of the Regulation, including provisions not listed in Article 99(4). The European Data Protection Supervisor can fine Union institutions, agencies and bodies in scope.

Role therefore matters. A provider of a GPAI model faces a different enforcement path from an enterprise that only uses such a model. Deciding which role applies to each AI use is the first step in sizing exposure. This guide does not forecast how or when any authority will act.

Testing whether current controls would hold up

The Act's legal duties are set out in the Regulation. The following points are practical judgments about useful evidence, not requirements stated in the Act.

Start with Article 5. Because prohibited practices carry the highest fine tier and already apply, an enterprise can test whether it can identify uses that could fall within the prohibitions, such as harmful AI-based manipulation and deception or individual criminal offence risk assessment or prediction. The harder question is whether those uses can be stopped or flagged at the point of use, rather than caught in a periodic review. Where a use sits near an Article 5 boundary, a defined escalation and human approval path gives the organisation a documented decision point.

If a national authority asks what an AI system was permitted to do and what it did, two records are useful: the policy that applied at the time, and a durable record of the interactions and actions that policy governed. Assembled after the fact from application logs, these are often incomplete or inconsistent. For AI literacy, which has applied since 2 February 2025, keep dated records of the measures taken and who they covered. For GPAI, confirm whether the organisation is a provider or a user of a model, since fine exposure follows from that.

A useful exercise is to take one AI use case and ask whether the evidence could be produced within a short response window, and whether it would show permitted behaviour and actual behaviour together.

Where runtime policy enforcement fits

Trussed enforces policy in the execution path of AI interactions and tool calls across models, agents, tools, MCP servers, AI sites and coding assistants. Each interaction can be allowed, redacted, blocked, rate limited or escalated for human approval, and this is enforced at runtime, not reviewed after the fact. The enforcement itself generates continuous audit-ready evidence, mapped to frameworks including the EU AI Act, so a regulated enterprise can show what its AI was allowed to do and what it did. Deployment is self-managed in the customer cloud, with configurable retention and data residency.

This addresses one part of the problem: operational control and evidence for obligations already in force. It does not deliver compliance on its own. Role determination, AI literacy programmes and legal assessment of Article 5 boundaries remain with the organisation.