Strasbourg: The AI Liability Directive, parked since spring 2024 amid Council resistance and a parallel push to finish the AI Act, is creeping back onto the Parliament’s Legal Affairs agenda. JURI shadow rapporteurs met in restricted session this week to map out a revised compromise text that would carry the file through trilogue before the autumn cliff. The Commission’s withdrawal threat, first floated in the 2025 work programme review, has been quietly dropped, and several legal services have begun preparing fresh impact data for the next inter-institutional reading.
What changed is the shape of the AI Act fallout. With high-risk system rules now applying since February 2026 and the General-Purpose AI code of practice in force, victims of AI-driven harm still have no harmonised civil route to compensation. National courts, particularly in Germany and the Netherlands, have begun pushing causality questions back through preliminary references, and the Commission’s services warn that the patchwork of fault-based and strict-liability regimes risks splintering the single market for high-risk applications. JURI’s draft text, leaked to specialised press this morning, retains the original rebuttable presumption of causality but narrows its scope to high-risk systems already covered by the AI Act, with a six-month transposition cliff after entry into force.
The disclosure obligation, the most contested chapter, survives in a tightened form. National judges would retain the power to order producers of high-risk AI systems to release technical documentation in proceedings, but only after a preliminary plausibility test and with reinforced trade-secret safeguards. Germany’s Free Democrats and the Renew shadow had pressed for a narrower disclosure window, and the new draft answers them halfway. The Permanent Representatives Committee will need to decide whether the Council general approach, frozen since the Belgian presidency, can absorb the changes without reopening the working party.
Compounding the timing pressure is the European Court’s pending judgment in Case C-441/25, which asks whether the existing Product Liability Directive recast captures damages from generative AI output. A finding for the claimant would compress the political space for Member States that have argued the new instrument is unnecessary. Several digital ministries have begun briefing capitals that a horizontal liability layer is preferable to a court-built doctrine read across 27 systems.
Industry has shifted, too. The European AI Alliance survey published in April showed 58 percent of mid-sized developers now favour a single liability frame over national divergence, citing insurance cost as the deciding factor. Munich Re’s pricing model, circulated to insurers in March, treats jurisdictional fragmentation as a 22 percent premium loading on professional indemnity products covering AI-assisted services. The smaller AI scale-ups represented in EU Inc circles have been the loudest converts, arguing that legal predictability matters more than the marginal cost of a presumption.
The Spanish presidency picks up the file from the Polish team on 1 July with a written mandate to reopen Council working party talks. Madrid’s negotiators are reportedly aiming for a partial general approach at the September JHA Council, leaving disclosure carve-outs for the final mile of trilogue. The Commission has confirmed it will not table a fresh proposal, preferring to let the institutions converge on the existing text.
Civil-rights coalitions are pushing back on the narrowed scope. EDRi’s brief to JURI warns that excluding low-risk and limited-risk systems strands most claimants in the same gap the file was meant to close. Consumer organisations in BEUC’s circle have asked for at least a review clause anchored to 2030. The political cost of a narrower text may yet outweigh the timing benefit, and Strasbourg is bracing for amendment fights through the summer recess. The autumn cliff is closer than it looks.
