AI Regulation Tracker / Enforcement
The EU AI Office Gets Real Enforcement Power Over Foundation Models on 2 August 2026
Regulatory summary: On 2 August 2026 the European Commission's enforcement powers over providers of general-purpose AI (GPAI) models enter into application under the EU AI Act. From that date the Commission, operating through the AI Office, can request documentation and information, conduct evaluations of GPAI models, require providers to take mitigation measures.
From 2 August 2026 the European Commission, acting through its AI Office, can compel documentation from general-purpose AI providers, run its own model evaluations, order mitigation or withdrawal, and fine providers up to 15 million euro or 3 percent of worldwide turnover. Models trained above 10 to the 25th power in floating point operations carry extra duties and are the likely first targets.
Key takeaways
- The obligations for GPAI providers have existed since 2 August 2025. From 2 August 2026 the Commission gains the operative power to enforce them. It can compel documentation and information, evaluate models directly, order mitigation measures, restrict or withdraw a model, and levy fines under Article 101. This converts a set of duties into a supervised, penalty-backed regime with the AI Office as the central enforcer.
- Foundation-model providers selling into or reaching the EU; enterprises embedding GPAI in products; and, downstream, professional-service teams in contract, compliance, procurement, and audit who now have to verify that the GPAI models inside their tools carry the required provider documentation.
- Status: Obligations in force since 2 August 2025.
- List every GPAI model your organization builds, embeds, or resells; for each, obtain or request the Article 53 documentation and confirm systemic-risk status; and write provider-compliance representations into procurement and vendor contracts before 2 August 2026.
| Date | Jurisdiction | Rule | Affected professionals | Status or effective date |
|---|---|---|---|---|
| 2026-07-09 | European Union | The obligations for GPAI providers have existed since 2 August 2025. From 2 August 2026 the Commission gains the operative power to enforce them. It can compel documentation and information, evaluate models directly, order mitigation measures, restrict or withdraw a model, and levy fines under Article 101. This converts a set of duties into a supervised, penalty-backed regime with the AI Office as the central enforcer. | Foundation-model providers selling into or reaching the EU; enterprises embedding GPAI in products; and, downstream, professional-service teams in contract, compliance, procurement, and audit who now have to verify that the GPAI models inside their tools carry the required provider documentation. | Obligations in force since 2 August 2025. Commission enforcement powers, including fines, enter into application on 2 August 2026 per the official guidelines page. A voluntary GPAI Code of Practice supports compliance in the interim. |
Frequently Asked Questions
Are GPAI obligations new on 2 August 2026?
No. The substantive obligations for GPAI providers applied from 2 August 2025. What enters into application on 2 August 2026 is the Commission's power to enforce them, including through fines, as confirmed on the Commission's GPAI guidelines page.
How much can a GPAI provider be fined?
Under Article 101, the Commission may fine a GPAI provider up to 3 percent of its total worldwide annual turnover in the preceding financial year or 15 million euro, whichever is higher.
What makes a model a systemic-risk model?
Under Article 51, a model is presumed to carry systemic risk when its cumulative training compute exceeds 10 to the 25th power floating point operations, or when the Commission designates it based on the Annex XIII criteria. Such models face extra testing, risk-assessment, and incident-reporting duties.
Who enforces the GPAI rules?
The European Commission, acting through the AI Office, enforces the GPAI model layer centrally at EU level. It can request documentation, evaluate models, order mitigation or withdrawal, and impose fines.
We only use a foundation model, we do not build one. Does this affect us?
You are not the fined party under Article 101, but you depend on a regulated component. Obtain the provider's technical documentation and training-data summary, confirm systemic-risk status, and add provider-compliance clauses to your contracts before 2 August 2026.
Internal links
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.