The EU AI Act's first enforcement deadline arrives August 2, 2026, and the compliance picture for General Purpose AI model providers is more chaotic than the regulation's supporters admit. The deadline triggers new obligations for anyone operating a GPAI model above the systemic risk threshold—broadly, models trained on compute exceeding 10^25 FLOPs. That threshold captures OpenAI's GPT-4 class and above, Google's Gemini family, Anthropic's Claude models, and Meta's Llama training runs. The companies that built the models driving the current AI wave.

The requirements at this stage are technical: providers must document training data sources, model architecture choices, capability evaluations, and known limitations. They must submit to the EU's AI Office a summary of the training data used, including a copyright assessment. They must have a written policy for handling downstream copyright takedown requests. These sound manageable until you try to operationalize them—which is exactly what most providers are discovering right now.

Who's Actually Ready

The honest answer: almost nobody. OpenAI has a team working on documentation, but the company has not published a training data summary that meets the regulatory specification. Google published a technical report for Gemini that doesn't cover the copyright assessment requirement. Anthropic is further ahead internally but hasn't published externally in a form the AI Office would accept.

What exists in the gap is a mix of placeholder compliance—published summaries that say very little while technically existing—and genuine non-compliance. The AI Office has been clear that it will accept incomplete submissions during the initial period, but "during the initial period" is doing a lot of work in that sentence, and the definition of "initial period" is itself subject to ongoing interpretation.

The Fine Structure Actually Matters

The penalties get the attention—up to €35 million or 7% of global annual turnover for GPAI violations—but the more material risk for most providers is operational pause. The AI Act allows the Commission to require providers to halt model availability in the EU market pending investigation of non-compliance. For a company deriving meaningful EU revenue from a flagship model, a market halt is a worse outcome than the fine.

What August 2026 actually marks is the beginning of an enforcement conversation between the AI Office and providers, not a cliff after which penalties automatically apply. The regulation is designed to work through dialogue and remediation in its early phases. That doesn't mean fines aren't coming—it means they're preceded by a process that most companies haven't fully staffed for.