EU AI Act deadline forces model vendors to lock plans this week as Brussels readies final guidance
Enzo Reno ·
The European Commission’s AI Office is moving to finalize codes of practice for general purpose AI this month in Brussels, and model providers are locking compliance decisions this week ahead of the EU AI Act’s broad applicability in early August 2026. The outcome will shape how OpenAI, Google, Anthropic, Meta, and Mistral configure model disclosures, evaluations, and incident reporting for customers across the bloc.
This matters now because enterprise buyers face near term changes to APIs, logging, and documentation that affect procurement and risk approvals. Gartner has projected that by 2026, more than 80 percent of enterprises will have used generative AI APIs and models, up from less than 5 percent in 2023, which raises the operational impact of any last minute compliance shifts.
What to watch
The AI Act introduces obligations for general purpose AI and a subcategory for models with systemic risk, with a compute proxy of around 10 to the power of 25 floating point operations cited in the final compromise text published in 2024, subject to Commission updates. The AI Office, which the Commission said would launch with about 140 staff, is expected to issue final guidance and codes of practice to standardize model cards, evaluation disclosures, and security reporting in the coming weeks.
Enforcement will involve the Commission for systemic risk models and national authorities for most use cases. The law sets penalties up to 35 million euros or 7 percent of global annual turnover for the most serious infringements, according to the published Act, with lower tiers for other violations and for providing incorrect information.
Vendors are preparing EU specific documentation and feature toggles, according to customer advisories and public statements. OpenAI, Google, Anthropic, and Meta have published model cards and watermarking or provenance plans aligned with commitments made to the White House in 2023, and they will need to align those artifacts with EU templates, including provenance signals that reference standards such as C2PA.
Enterprise impact
Cloud platforms will be central to compliance. Microsoft has said more than 65 percent of the Fortune 500 used Azure OpenAI Service by early 2024, and it offers an EU Data Boundary for storage and processing. AWS has announced a European Sovereign Cloud program, and Google Cloud has sovereign controls with partners including T Systems and Telecom Italia, all relevant to audit trails, dataset governance, and access controls required by the Act.
Member states will coordinate through the European AI Board, and the Commission’s AI Office will supervise systemic risk designations and cross border issues. France and Germany have backed domestic champions such as Mistral and Aleph Alpha, and they pressed during negotiations for lighter obligations on open models, an approach that will be tested as guidance lands.
For the wider tech sector, this week is a test of whether global model providers converge on an EU first compliance baseline or fragment their offerings by region, and whether compliance engineering becomes a default product feature for AI services sold worldwide.