AI Act pushed to December 2027: for regulated sectors the real issue isn't the reprieve, it's the 2 August 2026 transparency deadline

The Digital Omnibus package, approved by the European Parliament in June 2026, resets the AI Act timeline. Annex III high-risk systems — workforce management, hiring, scoring, educational technology — see their compliance deadline pushed from 2 August 2026 to 2 December 2027, a sixteen-month reprieve. Annex I safety components, which directly affect industrial machinery and medical devices, slip from 2 August 2027 to 2 August 2028. The official rationale is pragmatic: the harmonized technical standards were not ready, making compliance materially impossible within the original timeframe.

This easing comes with a narrower definition of "high-risk": automation tools aimed at efficiency, performance optimization or user convenience no longer trigger the intensive regime; only systems whose failure directly threatens physical safety or fundamental rights remain in scope. In parallel, simplified compliance tiers and regulatory sandboxes are opening for mid-market firms (up to 750 employees and €150M revenue). For many industrial projects, the expected regulatory burden genuinely decreases.

The trap lies in the gap between the media message — "Europe is loosening the rules" — and a deadline that does not move. Article 50 on transparency still applies on 2 August 2026: every customer-facing AI interaction and every generated synthetic media must be clearly identified as such, and watermarking of already-deployed generative content is extended to 2 December 2026. The penalties are not symbolic: up to €15M or 3% of global turnover. For a large aerospace, energy or defense group, this cross-cutting obligation covers customer assistants, internal copilots exposed to third parties and any automated content production.

Cardan-AI analysis: the right reading for industrial leadership is to separate two horizons. In the short term, an Article 50 compliance effort — inventory of AI touchpoints, labeling, governance of synthetic content — to be completed before 2 August 2026, without delay. In the medium term, use the sixteen-month high-risk reprieve not to stall but to build the governance layer (logging, human oversight, vendor documentation) that will still be required in 2027-2028. The reprieve rewards those who build, not those who wait. We support this dual framing, from compliance inventory to agent governance architecture.