Several shadow rapporteurs for the AI ‘omnibus’ (text revising the regulation on artificial intelligence) have published their amendments to the draft report drawn up by Michael McNamara (Renew Europe, Irish) and Arba Kokalari (EPP, Swedish) (see EUROPE 13802/11).
The original text contained only those details on which the two rapporteurs from the two competent committees, LIBE and IMCO, had been able to agree. The amendments from the various shadow rapporteurs are expected to add grist to the mill in the negotiations, which promise to be tough and highly political (see EUROPE 13784/5).
Axel Voss, shadow rapporteur for the EPP in the LIBE Committee, has posted his amendments online. In favour of the ‘stop-the-clock’ measure (the postponement of obligations for high-risk AI models), he removes the Commission’s flexibility mechanism for bringing forward the entry into force of these obligations and proposes fixed dates: 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems. These dates are in line with those proposed by the Council in its compromise (see EUROPE 13794/8).
The German politician also wants to restrict the ‘high-risk’ classification of AI systems that carry out profiling to cases where they are “likely to significantly influence the outcome of decision-making”.
He also suggests setting a threshold for fine-tuning AI models, to prevent even the smallest ‘refinement’ from representing a “substantial modification” and tipping some operators into the category of model providers.
He also wants to incorporate a precise, explicit and detailed definition for Open Source models in order to create legal certainty throughout the EU.
Svenja Hahn (German), shadow rapporteur for the Renew Europe group in the IMCO Committee, has also published her amendments. She wants to exclude from the scope of the text “industrial AI systems already covered by existing sectoral legislation, as well as systems that are purely internal to companies or B2B with no impact on consumers”.
She is also in favour of the use of certain personal data by AI “to detect bias and improve security within the clear standards of the GDPR”. “If we want European AI based on European values, we need to enable training based on European data”, she says.
She also wants to give generative AI developers 12 months to adapt their models to the legislation, instead of the 6 months proposed by the Commission.
For The Left, Leïla Chaibi (French), the shadow rapporteur in the IMCO Committee, is reinstating in her amendments the obligation to register in the European database for providers of AI systems that are not considered to be ‘high-risk’.
She also removes the possibility for AI models to use certain personal data, as well as many of the flexibilities offered to SMEs and SMCs (small and medium-sized enterprises).
As for the compliance deadlines for ‘high-risk’ systems, The Left’s amendments set them at 1 December 2026 for Annex III systems and 1 December 2027 for Annex I systems. (Original version in French by Isalia Stieffatre)