Europe’s AI Act Is Now Law: What the August 2 Deadline Means for General-Purpose AI Providers — and Why It Changes Everything

As of August 2, 2025, the European Union’s AI Act becomes legally enforceable for general-purpose AI models (GPAI), marking a historic moment in global technology governance. No longer a theoretical framework or an aspirational policy, the AI Act is now binding law, reshaping how artificial intelligence is developed, distributed, and governed across the EU — and beyond.

For companies building, hosting, or deploying GPAI models, the message is clear: compliance is no longer optional. The rules apply to all providers, regardless of where they are based, if their models reach users within the EU (European Commission, 2025a).

A Turning Point for General-Purpose AI

GPAI models — those capable of generating text, images, video, or audio across multiple domains — are no longer seen as experimental or peripheral. They are now classified as high-impact digital infrastructure with systemic implications for markets, media, education, and democracy (Agostini, 2025; European Commission, 2025a).

To ensure their safe and responsible use, the Act introduces a series of mandatory obligations. Providers must create and maintain technical documentation detailing a model’s architecture, safety features, and performance boundaries. They are required to publish a summary of the training datasets, indicating sources, representativeness, and dataset integrity. And they must ensure compliance with EU copyright rules, especially under the Digital Single Market directive (European Parliament & Council, 2019).

For high-impact models — specifically those trained with over 1⁰²³ FLOPs and possessing multimodal generative capacity — the obligations extend further. Providers must conduct ongoing risk assessments, implement continuous monitoring mechanisms, enable independent third-party audits, and notify regulators of any systemic incidents (European Commission, 2025a; Bloomberg Law, 2025).

These obligations also apply to non-EU companies. If a model is made available to European users via API, downloads, or embedded services, the provider must appoint an EU-based representative and adhere to all regulatory requirements (European Commission, 2025b).

It’s Not Just About Developers

The AI Act also targets integrators and distributors — entities that embed GPAI into their own platforms or applications. These companies must verify upstream compliance, ensure that their use of the model does not introduce new risks, and disclose when AI-generated outputs are being presented to users (European Commission, 2025a). Failure to do so exposes them to the same penalties as original providers.

Supervision and enforcement are coordinated at the EU level. Member States are required to appoint national authorities and align with the newly established European AI Office, which will oversee cross-border cooperation via the European Artificial Intelligence Board (European Commission, 2025b).

Enforcement Comes with Teeth

The fines are substantial — and scalable. Companies face penalties of up to €35 million or 7% of global annual turnover for violating bans on prohibited practices. Noncompliance with obligations concerning high-risk models could result in fines up to €15 million or 3%, while submitting false or incomplete documentation may lead to penalties of €7.5 million or 1% (AP News, 2025).

A Voluntary Code — But Not a Loophole

To support early implementation, the EU introduced a voluntary Code of Practice for GPAI in July 2025. This Code offers templates for risk disclosures, data transparency, and copyright management, and while it does not replace legal requirements, it is intended to ease the compliance burden and signal good faith to regulators (Reuters, 2025). Google signed the Code. Meta did not (Financial Times, 2025b).

As Bloomberg Law (2025) notes, voluntary alignment may help reduce enforcement scrutiny in the early stages of rollout — but does not immunize companies from penalties if they fail to meet formal obligations.

What Defines a General-Purpose AI Model?

The Commission’s July 2025 guidance defines a GPAI model as one trained with compute greater than 1⁰²³ FLOPs and capable of producing multimodal outputs. Models used for narrow tasks like transcription or enhancement are excluded. However, the definition of “provider” includes any entity that makes the model available under its own brand, even if it did not develop it internally (European Commission, 2025a).

Entities that modify the model’s risk profile or functionality are also deemed providers under the law and assume full regulatory responsibility.

Some open-source models may qualify for limited exemptions, particularly if they are released under permissive licenses and not intended for commercial use. However, transparency and auditability are still expected (European Commission, 2025a).

Industry Pushback — and the EU’s Firm Response

In June 2025, more than 150 CEOs — including from Airbus, Carrefour, and BNP Paribas — signed a letter urging the EU to delay enforcement by two years, citing concerns about legal clarity and global competitiveness (Financial Times, 2025a). The Commission rejected the request.

Instead, it reiterated its commitment to a phased and collaborative enforcement strategy, particularly for SMEs and open research projects.

Conclusion: Why This Is About More Than Compliance

Europe is not just regulating artificial intelligence — it is redefining digital market governance.

The rationale is both pragmatic and strategic. General-purpose models now underpin systems that shape public discourse, automate judgment, and influence economies (Agostini, 2025). Yet many of these models are built on data that is poorly documented, hard to verify, and trained in conditions that escape institutional oversight (Bloomberg Law, 2025; European Commission, 2025a). As these models become increasingly embedded in decision-making infrastructure, the risks they pose — discrimination, misinformation, copyright violation — cannot be left to self-regulation alone (Reuters, 2025; Financial Times, 2025b).

The EU has chosen to act — not react. The AI Act enforces a framework of transparency, responsibility, and traceability, providing not just legal clarity but a governance blueprint for the global digital economy.

As I previously argued (Agostini, 2025), the AI Act is more than a compliance checklist. It’s a strategic architecture for trust. Companies that adapt early will not only reduce regulatory risk — they’ll build long-term legitimacy.

References

Agostini, M. (2025, July 25). Europe just set the rules for general-purpose AI — Here’s what it means for your business. Medium. https://medium.com/@tarifabeach/europe-just-set-the-rules-for-general-purpose-ai-heres-what-it-means-for-your-business-2e721b4cc2ed

AP News. (2025, July 19). Europe’s AI Act enters enforcement phase, with fines up to 7% of global turnover. https://apnews.com

Bloomberg Law. (2025, July 22). A lawyer’s guide to the EU AI Act. https://pro.bloomberglaw.com/insights/technology/a-lawyers-guide-to-the-eu-ai-act

European Commission. (2025a, July). Guidance on GPAI model classification and provider obligations. Publications Office of the European Union. https://digital-strategy.ec.europa.eu/en/policies/ai-act

European Commission. (2025b). AI Act: Regulatory framework overview. https://digital-strategy.ec.europa.eu/en/policies/ai-act

European Parliament & Council. (2019). Directive (EU) 2019/790 on copyright and related rights in the Digital Single Market. Official Journal of the European Union. https://eur-lex.europa.eu

Financial Times. (2025a, June 30). EU CEOs urge delay to AI Act enforcement. https://www.ft.com/content/a825759e-aec8-4184-bc73-f604f169204c

Financial Times. (2025b, July 10). Google signs EU AI code of conduct as Meta holds back. https://www.ft.com/content/97ee867c-64c2-44ee-a519-36ca670ed565

Reuters. (2025, July 18). EU launches AI Code of Practice to help companies meet Act requirements. https://www.reuters.com

AIAct, #GPAI, #GeneralPurposeAI, #AIGovernance, #AICompliance, #AIEthics, #AIRegulation, #AITransparency, #AITrust, #AISafety, #AIAudit, #RiskManagement, #DigitalMarketRegulation, #ArtificialIntelligence, #EUAIAct, #OpenSourceAI, #AIOversight, #AIProviders, #AIAuditTrail, #AIEnforcement, #TechRegulation, #AIPolicy, #AIInfrastructure, #AIModels, #AIDocumentation