The EU can now fine general-purpose AI providers. We put the one-year odds of a first decision at 35%

The narrow event is not that the entire AI ActAI ActThe European Union regulation that sets rules for certain AI systems and general-purpose AI models. [AI Act Explorer](https://ai-act-service-desk.ec.europa.eu/en/ai-act-explorer) suddenly appeared. The obligations for providers placing general-purpose models on the EU market have applied since 2 August 2025. What changed on 2 August 2026 is the Commission's ability to enforce those obligations, including through Article 101Article 101The part of the AI Act that authorises Commission fines for specified intentional or negligent conduct by general-purpose AI providers. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101) fines. [Commission guidelines](https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpaigpaiThe abbreviation for general-purpose artificial intelligence.-providers) [AI OfficeAI OfficeThe Commission function entrusted with implementing, monitoring and enforcing the EU rules for general-purpose AI providers. [Article 88](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-88) FAQ](https://ai-act-service-desk.ec.europa.eu/en/faq?faq_category_id=69)

There is an important timing wrinkle. Commission Implementing Regulation 2026/1755 sets the detailed arrangements for model evaluations, opening and closing fining proceedings, preliminary findingspreliminary findingsThe Commission's written account of the suspected infringement before it adopts a final fine decision. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101), access to the file and limitation periods. It was published in the Official Journal on 21 July 2026 and takes effect on 10 August 2026. [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng)

Article 101 covers four routes. The Commission may fine a providerproviderA person or organisation that develops, or has developed, an AI model and supplies it under its own name or trademark. [Article 3](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-3) for infringing relevant AI Act provisions, failing to comply with a document or information requestinformation requestA demand for documents or other information needed to assess whether a provider is following the law. [Article 91](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-91) or supplying incorrect, incomplete or misleading information, failing to comply with a measure requested under Article 93, or failing to provide access needed for a model evaluationmodel evaluationA regulatory examination of a model that may involve testing and access through technical interfaces or source code. [Article 92](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-92) under Article 92. The conduct must be intentional or negligent. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)

That means a first fine would not automatically be a verdict that a model was dangerous. It could instead concern regulatory cooperation. The distinction is similar to a tax authority penalising an inadequate filing: the penalty says the legal process was not followed, even if it does not decide every underlying question about the business. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)

The maximum is 3% of the provider's total worldwide turnover in the preceding financial year or €15 million, whichever is higher. The Commission must consider the nature, gravity and duration of the infringement, proportionality, appropriateness and relevant commitments when setting the amount. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)

The Commission can use information requests, model evaluations and requested corrective measures before opening a fining proceeding. It may also order an urgent interim measureinterim measureA temporary urgent restriction that can be ordered before a fining proceeding when the Commission initially finds an infringement and a risk of serious damage. [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng) before opening proceedings if it makes an initial finding of an infringement and identifies a risk of serious damage to health, safety or another protected public interest. [Articles 91 to 93](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-91) [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng)

For a final fine, Article 101 requires preliminary findings and an opportunity for the provider to be heard. The implementing regulation gives the provider at least 21 days to submit written observations and evidence. Upon request, the provider also receives access to the Commission's file under confidentiality safeguards. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101) [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng)

The regulation does not impose a one-year deadline for completing a case. Instead, it generally gives the Commission five years from the conduct to adopt a fine, with special rules for continuing conduct and actions that interrupt the clock. That does not mean a case will take five years, but it shows that the statutory timetable is much wider than our forecast window. [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng)

We looked for recent EU digital regimes in which the Commission directly supervises large technology companies, can investigate non-compliance and can impose turnover-based fines. The Digital Markets Act and the Digital Services Act are the closest available analogies. They are not a statistical sample and they regulate different conduct, but they provide real administrative timelines.

Under the Digital Markets Act, the first six designated gatekeepers had to comply fully on 7 March 2024. The Commission opened its first non-compliance investigations on 25 March, 18 days later. It adopted the first fines on 23 April 2025, fining Apple €500 million and Meta €200 million. The compliance-date-to-fine interval was 412 calendar days. [DMADMAThe Digital Markets Act, an EU law imposing obligations on designated large technology platforms called gatekeepers. compliance date](https://digital-markets-act.ec.europa.eu/designated-gatekeepers-must-now-comply-all-obligations-under-digital-markets-act-2024-03-07_en) [DMA investigations](https://digital-markets-act.ec.europa.eu/commission-opens-non-compliance-investigations-against-alphabet-apple-and-meta-under-digital-markets-2024-03-25_en) [DMA decisions](https://digital-strategy.ec.europa.eu/en/news/commission-finds-apple-and-meta-breach-digital-markets-act)

For the Digital Services Act, the comparable start is not its general application date of 17 February 2024. The first directly supervised very large platforms and search engines were already subject to the regime in August 2023, and the Commission's enforcement page dates its investigatory record from 25 August 2023. The first DSADSAThe Digital Services Act, an EU law governing online intermediaries and imposing additional duties on the largest platforms and search engines. non-compliance decision came on 5 December 2025, when the Commission fined X €120 million. That produces an 833-day interval. [DSA enforcement framework](https://digital-strategy.ec.europa.eu/en/policies/dsa-enforcement) [DSA decision](https://digital-strategy.ec.europa.eu/en/news/commission-fines-x-eu120-million-under-digital-services-act)

Both precedents missed the one-year mark, which pulls our forecast below 50%. We do not treat 412 and 833 days as a mechanical average because two different laws cannot reveal a reliable average enforcement time. Their value is directional: a Commission fine can take longer than the headline start date suggests.

The AI regime did not begin from a blank file on 2 August 2026. The model-provider obligations had already applied for one year, and the AI Office says it had received information and held technical compliance dialogues with providers during that period. It describes those dialogues as its preferred first tool but says compulsory powers may be used when dialogue is insufficient. [AI Office FAQ](https://ai-act-service-desk.ec.europa.eu/en/faq?faq_category_id=69)

Capacity has also increased. The Associated Press reported on 31 July 2026 that the AI Office was adding 38 people to monitor AI companies. That is evidence of greater enforcement capacity, although headcount alone does not prove that a case is ready or that a decision will be adopted within a year. [Associated Press](https://apnews.com/article/eu-ai-regulation-deepfakes-hacking-f4fcee1f9750e2b32cdf26ad73ee5ec2)

The Commission's public Code of PracticeCode of PracticeA voluntary framework that providers can follow to help demonstrate compliance with the AI Act's model obligations. [Commission Code page](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai) page listed 21 full signatories and identified xAI as a signatory only to the Safety and Security chapter on 31 July 2026. The Code is voluntary, the page warns that confirmed signatories may not appear immediately, and providers may use other ways to demonstrate compliance. The list therefore shows an observable group, not the total population exposed to Article 101. [Commission Code page](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai)

The breadth of Article 101 also raises the chance of a qualifying decision. A fine can address an underlying breach, but it can also address incomplete or misleading information, failure to comply with a requested measure or failure to provide evaluation access. The first case therefore need not depend on proving the most technically difficult allegation about a model's real-world safety. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)

The procedure contains real defence rights and no fast-track deadline. A provider can answer preliminary findings, submit evidence and request access to the file. Complex questions about model classification, provider identity, technical access, confidentiality and systemic risksystemic riskA risk from a highly capable or widely consequential model that could significantly affect health, safety, security, fundamental rights or society across the EU. [Commission guidelines](https://digital-strategy.ec.europa.eu/en/faqs/guidelines-obligations-general-purpose-ai-providers) could lengthen a case even when the Commission is motivated to move. [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng) [Commission guidelines](https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers)

The regulated population is also staggered. Providers of models placed on the EU market before 2 August 2025 have until 2 August 2027 to comply. Much of the older model stock therefore reaches its compliance deadline only at the forecast cutoff, limiting the time available for a pre-cutoff case based on those obligations. [Article 111](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-111) [Commission guidelines](https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers)

Finally, the AI Office says it will continue and intensify technical compliance dialogues. Dialogue can expose a breach, but it can also produce correction without a monetary decision. Article 93 separately allows requested measures and binding commitments, giving the Commission outcomes short of a fine. [AI Office FAQ](https://ai-act-service-desk.ec.europa.eu/en/faq?faq_category_id=69) [Article 93](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-93)

We did not use share-price moves in this forecast. No particular provider is the subject of the question, and a stock move after regulatory news would measure investors' changing expectations, not the eventual compliance bill.

The economic effecteconomic effectA change to costs, revenue, operations or product availability, such as compliance spending, a fine or a required model withdrawal. would be the fine actually imposed, legal and compliance spending, operating changes or a required restriction, withdrawal or recall. Article 93 authorises those corrective measures, while Article 101 sets the fine ceilingfine ceilingThe highest fine the law permits, not the amount a regulator must or probably will impose.. Until one of those actions occurs, the maximum penalty is exposure rather than a realised cost. [Article 93](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-93) [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)

We assign a 35% probability to at least one Commission-adopted Article 101 monetary fine by the end of 2 August 2027.

The 35% forecast balances a prepared regulator, a year of prior compliance work and several possible grounds for a fine against a formal procedure, defence rights, a staggered regulated population and two close precedents that took more than one year. It is a structured judgment rounded to the nearest five percentage points, not a prediction generated by a large historical dataset.

• The Commission publicly opens an Article 5 proceeding aimed at a possible Article 101 decision. [Implementing Regulation 2026/1755](https://eur-lex.europa.eu/eli/reg_impl/2026/1755/oj/eng)
• The Commission sends preliminary findings to a provider. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)
• A provider publicly fails to meet an information, access or corrective-measure demand covered by Article 101. [Article 101](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101)

• Public cases are resolved through dialogue, corrective measures or binding commitments without a fine. [Article 93](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-93)
• No formal proceeding is visible by early 2027, leaving less time for preliminary findings and the provider's response.
• Further legislation changes the applicable duties, procedure or enforcement timetable.