Analysis AI Compliance

EU AI Act: the updated enforcement timeline after the Digital Omnibus

The EU AI Act can apply to your organisation even if you do not build AI: it covers organisations that use AI systems in the EU, and providers and deployers outside the EU whose AI output is used in the EU. In July 2026, the Digital Omnibus amendments pushed Annex III high-risk AI obligations from August 2026 to December 2027. Article 50 transparency obligations covering chatbot disclosure, deepfake labelling, and synthetic content marking took effect on 2 August 2026. This article sets out the revised timeline and what your organisation needs to have in place.

5 August 2026
Updated 19 November 2026
8 min read
Key takeaways
  • Article 50 transparency obligations have applied since 2 August 2026: providers must tell people when they are interacting with AI and mark synthetic content in a machine-readable format; organisations that publish deepfakes must disclose that the content is generated or manipulated
  • Generative AI systems on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking to their output
  • Annex III high-risk AI obligations were delayed to 2 December 2027 by the Digital Omnibus, giving organisations 16 additional months of preparation time
  • High-risk categories include hiring, credit scoring, biometrics, critical infrastructure, education, and law enforcement
  • Organisations that use AI (deployers) carry distinct legal obligations, separate from those of their AI vendors (providers)
  • Fines reach up to 35 million euros or 7% of global turnover for the most serious violations

Where the enforcement timeline stands

The EU AI Act entered into force on 1 August 2024. Enforcement has rolled out in three waves. The first, in February 2025, banned a defined set of AI practices under Article 5: manipulative techniques and AI that exploits vulnerabilities, social scoring that leads to detrimental or disproportionate treatment, whether carried out by public bodies or private companies, systems that scrape facial images to build recognition databases and, as a rule, real-time remote biometric identification in publicly accessible spaces for law enforcement. That last ban has narrow exceptions, such as searching for abduction victims or missing persons, preventing an imminent threat to life or a terrorist attack, and locating suspects of serious crimes, and each use normally needs prior authorisation from a judicial or independent administrative authority.

The second wave arrived in August 2025. Providers of general-purpose AI models such as OpenAI's GPT-4, Anthropic's Claude, and Google's Gemini became subject to new transparency and copyright compliance requirements. Those obligations fall on the model developers, not on the businesses that access those models via API.

The third wave took effect on 2 August 2026, splitting into two parts. Article 50 transparency obligations applied on schedule: providers must make sure systems that interact with people tell them they are dealing with AI (unless that is obvious) and that synthetic content is marked in a machine-readable format; deployers must disclose deepfakes, and AI-generated text published to inform the public on matters of public interest, subject to the exceptions in the Article. Our Article 50 guide covers these obligations in detail. Generative AI systems already on the market before that date have until 2 December 2026 to add machine-readable marking. The European AI Office gained its full penalty enforcement powers over GPAI providers on the same date.

Annex III high-risk AI obligations moved on a different track. In July 2026, the EU adopted the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026), shifting the compliance deadline for standalone high-risk systems (those not embedded in regulated products) from August 2026 to 2 December 2027. High-risk AI embedded in Annex I regulated products moved from August 2027 to August 2028. The harmonised technical standards required to operationalise compliance were not finalised in time.

August 2024
AI Act enters into force
Regulation published and binding. Two-year transition period begins.
February 2025
Prohibited AI systems banned
The Article 5 prohibitions apply, including social scoring, manipulative AI and, subject to narrow exceptions, real-time remote biometric identification in publicly accessible spaces for law enforcement.
August 2025
GPAI model obligations apply
Providers of general-purpose AI models must comply with transparency and copyright rules.
August 2026
Article 50 transparency obligations in force
Chatbot disclosure, synthetic content labelling, and deepfake marking become enforceable. European AI Office gains full penalty powers over GPAI providers.
2 December 2026
Grace period ends for generative AI marking
Generative AI systems placed on the market before 2 August 2026 must add machine-readable marking to their output.
2 December 2027: DEADLINE
High-risk AI (Annex III) must comply
Delayed from August 2026 by the Digital Omnibus. Standalone Annex III high-risk AI systems (employment, credit scoring, biometrics, education, critical infrastructure) must meet full conformity requirements.
August 2028
Annex I product-embedded AI
High-risk AI embedded as a safety component in regulated products (medical devices, machinery, vehicles) must comply. Delayed from August 2027 by the Digital Omnibus.
€35M
maximum fine for prohibited AI violations, or 7% of global annual turnover
Dec 2027
Annex III high-risk AI compliance deadline, moved from August 2026
8
Annex III risk categories covering AI used in hiring, credit, biometrics, and more

Is your AI use high-risk? What Annex III covers

Annex III of the EU AI Act lists eight categories of high-risk AI use. The list is broader than most organisations expect. You do not need to build machine learning models to fall under it. A third-party tool that automates or materially influences a decision in any of these categories brings you in scope as a deployer.

The eight categories are:

  • Biometric identification and categorisation of natural persons, including remote identification systems and emotion recognition tools
  • Critical infrastructure management, including AI used in electricity, water, gas, transport, and digital infrastructure operations
  • Education and vocational training, including AI that determines access to educational institutions or assesses students (automated exam scoring, admission ranking)
  • Employment, worker management, and access to self-employment, including CV screening, recruitment ranking, performance monitoring, and promotion or dismissal decisions supported by AI
  • Essential private services, including AI used in credit scoring, insurance risk assessment, life and health insurance underwriting
  • Law enforcement, including AI used for risk assessment of individuals, profiling, evidence evaluation, and crime analytics
  • Migration, asylum, and border control management, including risk assessment of visa applicants and detection of document authenticity
  • Administration of justice and democratic processes, including AI that assists courts in researching or applying the law

Employment and essential services catch more commercial organisations than any other category. If your HR platform ranks job applicants, your CRM scores customers for credit risk, or your performance management software assigns productivity ratings to workers, you are probably using a high-risk AI system under Annex III.

There are exceptions. Under Article 6(3), a system listed in Annex III is not high-risk when it poses no significant risk to health, safety or fundamental rights, for example when it performs a narrow procedural task or only improves the result of a human activity already completed. The exception never applies to systems that profile people, and a provider relying on it must document the assessment and still register the system.

The Act targets the use of AI in consequential decisions about people. Most affected organisations are buyers of third-party tools, not AI developers.

If you find a high-risk use, EU AI Act high-risk readiness sets out what those systems must have in place by December 2027, which duties fall on you as a deployer rather than on your AI vendor, and a four-week plan with a checklist of where to focus.

The fine structure

The EU AI Act uses the same tiered penalty model as GDPR: fines are the higher of a fixed euro amount or a percentage of global annual turnover. For SMEs and startups, the lower of the two amounts applies. The Act sets fines for organisations; any personal liability for directors depends on national law in each member state.

  • Prohibited AI violations (Article 5): up to 35 million euros or 7% of global annual turnover, whichever is higher
  • Breach of other operator obligations (providers, deployers and other operators, including high-risk obligations and the Article 50 transparency duties): up to 15 million euros or 3% of global annual turnover, whichever is higher
  • Incorrect information to authorities: up to 7.5 million euros or 1% of global annual turnover, whichever is higher

For a company with one billion euros in global revenue, the ceiling for breaching high-risk obligations is 30 million euros; the actual fine depends on the seriousness and duration of the breach and on cooperation with the authority. National supervisory authorities in each EU member state enforce the Act. GDPR precedent suggests the largest fines will target systematic failures and cases where individuals were harmed, but early enforcement decisions will set the tone.


How Cyvra helps with EU AI Act compliance

Cyvra works with deployers to identify which AI systems fall under Annex III, build the documentation those systems require, and establish the human oversight processes the regulation demands. Our focus is your obligations as a user of AI, separate from what your vendors owe.

  • AI system inventory: catalogue every AI tool in use across your organisation, including embedded AI features in software your IT team approved for other purposes
  • Annex III classification: assess each system against the eight high-risk categories and document the reasoning in a format regulators can audit
  • Vendor documentation review: confirm your AI providers have supplied the required technical documentation and conformity declarations, and identify where they have not
  • Human oversight design: build the processes and controls that high-risk AI deployments must have in place before December 2027
  • ISO 42001 alignment: map your AI Act compliance work to the ISO 42001 AI management system standard, producing a single audit-ready framework rather than two parallel compliance exercises

Talk to our AI practice or compliance team about what Article 50 requires now and how to prepare for the December 2027 Annex III deadline.

Frequently asked questions

Does the EU AI Act apply to my organisation if we are not an AI company?

It can. If your organisation uses AI in the EU, or its AI output is used in the EU, in Annex III categories such as HR, hiring, credit scoring, biometrics, or critical infrastructure, you are a deployer under the Act and carry legal obligations. You do not need to build or sell AI to be in scope. Most affected organisations are users of third-party tools, not AI developers.

What are the fines for breaching the EU AI Act?

Fines depend on the violation type. Using prohibited AI systems (those banned under Article 5) can reach 35 million euros or 7% of global annual turnover, whichever is higher. Breaching other operator obligations, including high-risk obligations and the Article 50 transparency duties, can reach 15 million euros or 3% of turnover. Giving incorrect or misleading information to authorities can reach 7.5 million euros or 1% of turnover. For SMEs and startups, the lower of the two amounts applies.

Does using ChatGPT or Claude via API make us subject to the GPAI provisions?

If you access a general-purpose AI model via API for your own internal use, you are a deployer, not a GPAI provider. GPAI obligations fall on the model developers: OpenAI, Anthropic, Google, and others. If you integrate a GPAI model into a product or system that other people use, your obligations increase. You then need to assess whether the resulting system qualifies as high-risk under Annex III.

Ryland Deakin
About the author
Lead Consultant, Cyvra · CISM · CompTIA Security+ · MCP

Ryland has delivered cybersecurity, compliance, and IT management programmes for regulated organisations across the UK and the Netherlands for over 20 years. View full profile

Talk to Cyvra

December 2027 is the Annex III deadline. Article 50 applies now. Start with your AI inventory.

Our AI and compliance practice helps you build your inventory, classify Annex III exposure, and confirm vendor documentation before enforcement begins.

Disclaimer: This article is for general informational purposes only and does not constitute legal, regulatory, or professional advice. Cyvra makes no warranty as to the accuracy or completeness of this content. Readers should seek independent legal advice appropriate to their specific circumstances. Cyvra accepts no liability for any loss arising from reliance on this content.