General-purpose AI obligations have applied since 2 August 2025, and the AI Office began publishing the Code of Practice signatory list from that date. A year on, the picture is a set of models operating under transparency and copyright commitments, and a smaller set carrying systemic-risk duties above the 10^25 FLOP threshold.
Signing the code bought signatories a presumption of good-faith compliance and a structured relationship with the AI Office. It did not exempt anyone from the underlying rules.
The change coming on 2 August 2026 is enforcement. The powers that let the AI Office act on the GPAI obligations switch on, and the Omnibus centralised that supervision at EU level, with national authorities retaining carve-outs for downstream deployment.
Analysis is original writing by the euaicompliant desk. Not legal advice — the binding text is on EUR-Lex.