EU AI Act transparency rules take effect August 2, 2026 — Commission consultation closes on draft guidelines, binding interpretations expected Q3
The EU AI Act's transparency provisions come into effect on August 2, 2026 — 70 days from this PM cycle. The Commission has opened consultation on draft guidelines for the transparency obligations, which will turn into binding interpretations during Q3. Companies developing or deploying GPAI models in the EU will need to disclose summary training-data information, downstream use restrictions, and high-risk-system risk-management documentation under the new rules.
The training-data disclosure obligation is the part US-headquartered labs have most actively pushed back on. OpenAI, Anthropic, and Google argue that a public-facing summary of training data exposes trade secrets and creates litigation surface for any included-but-disputed copyright owner. The Commission's draft guidelines are expected to require a structured-summary format rather than a raw inventory — but even that is an information asymmetry break vs the "black-box training set" position that has been the industry default.
The cross-Atlantic friction is visible in the contrast with the White House's March 20 National Policy Framework, which is a set of legislative recommendations rather than law. The EU has binding transparency rules effective in 70 days; the US has a framework that may or may not produce legislation by 2027. The divergence is what the "transatlantic divergence is real" blog framing this morning (AM cycle) was setting up. The next 70 days will determine whether US labs file early or push for delay; the EU has shown no appetite for either.
European Commission — AI Act regulatory framework → · AI Act EU — Up-to-date developments and analyses of EU AI Act → · AI Forest — AI Regulation News 2026 US EU UK Japan China Frameworks →