European lawmakers demanded clearer responsibility for harm caused by artificial intelligence, challenging the Commission’s decision to withdraw dedicated liability legislation. A debate on global AI governance also exposed a divide over Europe’s influence: whether stronger safeguards would help it lead, or whether investment and fewer barriers must come first.

International Partnerships Commissioner Jozef Síkela defended the EU’s approach of combining innovation, safeguards and international cooperation. “We firmly believe that appropriate safeguards do not stifle innovation,” he said. But lawmakers questioned whether existing rules were enough, who should enforce them and how Europe could shape technology largely developed elsewhere.

Who pays when AI causes harm?

“Stop letting Big Tech police itself,” said Alex Agius Saliba (S&D/MLT). He called the withdrawal of the AI Liability Directive “a big mistake” and urged the Commission to revive it. The proposed directive, introduced in 2022 and withdrawn in October 2025, sought to make it easier to pursue compensation for certain AI-related harm.

Kim Van Sparrentak (Green-EFA/NLD) used an everyday comparison. “If a phone charger explodes in your house, we never call it ‘innovative’,” she said. “We call it cheap, dangerous junk, and we hold the manufacturer liable for damages.” She demanded liability rules covering AI testing and development.

Fellow Green Sergey Lagodinsky argued that businesses also needed certainty about their responsibilities: “The companies need to know how much they would pay if something happens.”

If a phone charger explodes in your house, we never call it ‘innovative’. We call it cheap, dangerous junk, and we hold the manufacturer liable for damages. — Kim Van Sparrentak (Green-EFA/NLD)

Mr Síkela acknowledged those concerns but pointed to opposition among governments. “The Commission had proposed it, the Commission was in favour of it,” he said. There had been no agreement between Parliament and the Council, and none was expected soon.

He argued that the AI Act and revised product-liability legislation addressed some of the same concerns. The latter updates compensation rules for defective products, including software. Countries have until 9 December 2026 to implement it.

The commissioner offered no commitment to revive the directive. MEP Brando Benifei (S&D/ITA) nevertheless pressed the connection between global leadership and accountability at home. “Let’s lead, or we will need to wait for the next blog post of a tech CEO to know what is happening,” he said.

Europe needs rules, but also technology of its own

Arba Kokalari (EPP/SWE) challenged calls for further legislation when Europe already had the AI Act. “If Europe only focuses on regulating AI, no other country will follow us,” she said. Her priorities were easier access to investment, data and computing infrastructure, alongside fewer obstacles between research and the market.

The disagreement became direct when Leire Pajín (S&D/ESP) challenged that framing. Speaking in Spanish, she said the demand was for existing legislation to be applied, with ethical standards, oversight and transparency. Ms Kokalari agreed on enforcement but rejected additional bureaucracy. Kristian Vigenin (S&D/BGR) warned that “simplification must never become deregulation”.

The companies need to know how much they would pay if something happens. — Sergey Lagodinsky (Green-EFA/DEU)

There was broader agreement that European influence required investment as well as legislation. Eva Maydell (EPP/BGR) backed stronger enforcement alongside computing capacity, capital and a less fragmented single market. She also proposed a permanent dialogue between European leaders and technology experts, meeting every six months with concrete commitments.

Reinier Van Lanschot (Greens-EFA/NLD) made a similar case for building European alternatives: “Regulation is good, but not enough. To take charge of our digital future, we must build it ourselves.”

That shared concern extended to dependence on foreign suppliers. Markéta Gregorová (Greens-EFA/CZE) proposed stronger European testing, joint procurement and fallback arrangements for critical AI tools. “A tool that someone else can switch off is not our defence,” she said. These were lawmakers’ proposals, rather than measures agreed during the debate.

Deciding who joins the table

Michael McNamara (Renew/IRL) challenged the Commission’s emphasis on cooperation with like-minded partners. Global governance required engagement with China too, he argued. “We have to start talking to people that we feel threatened by. We have to start talking to people that we do not agree with.” Enforcing Europe’s own rules remained essential to making that international case credible.

We have to start talking to people that we feel threatened by. We have to start talking to people that we do not agree with. — Michael McNamara (Renew/IRL)

Lawmakers also differed over the model for cooperation. Mr Lagodinsky called for an AI restraint treaty drawing on arms control. Barry Andrews (Renew/IRL) questioned that analogy because AI was widely used and largely privately owned. He favoured a comparison with global health governance, including reporting and emergency-response arrangements when dangerous incidents occur.

International structures already exist. The UN established an independent scientific panel and a Global Dialogue on AI Governance in 2025, with the first dialogue held in July 2026. Those forums provide places to exchange evidence and discuss cooperation, rather than the binding treaty some lawmakers sought.

Mr Síkela promised more active engagement through the UN, G7, G20, OECD and AI safety bodies, alongside cooperation on testing, verification and early warning. “It is also intellectually honest to say that we do not know everything,” he said in closing. He did not endorse a new treaty or announce replacement liability legislation.