The European Parliament has adopted a resolution calling for any public decision assisted by artificial intelligence to remain under the effective control of a person, and for MEPs to keep a strict right of scrutiny over AI governance. The text also calls for a revision of the Charter of Fundamental Rights to enshrine new AI-related rights, a reform of European electoral law and an international convention on AI. It insists on the effective implementation of the European AI Act, which is not yet fully applicable. Parliament is asking, it is not setting a binding rule: this is a resolution, not a law.
AI is shaking up how administrations and institutions work: tools can prepare decisions that affect citizens, and automated systems could, according to Parliament, gradually capture choices that belong to elected officials. The text starts from the principle that AI governance is not just a technical question: it touches on public authority, democratic legitimacy and the balance between the European institutions. Parliament also sees it as a matter of digital autonomy and technological sovereignty for the Union.
A human always supervises AIThe deployment and use of AI within the European institutions must remain human-centred, with a person genuinely supervising each tool during its use. The principle of the human decision-maker must apply in all circumstances, so that democratic mandates do not silently pass to automated systems.
Right to human interventionEveryone should be able to obtain meaningful human intervention in decisions taken or supported by AI within public administrations. Parliament calls for this right to be recognised and genuinely guaranteed.
New rights in the CharterParliament calls for an amendment to the Charter of Fundamental Rights to add AI-related rights: protection against deepfakes, algorithmic transparency, digital self-determination, protection against algorithmic discrimination, and liability in the event of harm.
Electoral law and campaignsIt calls for a reform of European electoral law to regulate the use of AI in campaigns, political advertising and voter targeting, in order to make elections more resilient to disinformation.
Scrutiny of implementing actsThe growing use of delegated and implementing acts in the field of AI must not weaken Parliament's role as co-legislator. It calls for strengthened scrutiny mechanisms and regular coordination with the Commission and the Council, which could be formalised through an interinstitutional agreement.
Adopted by 405 votes to 133, with 60 abstentions, in an electronic roll-call vote.
In practice, nothing changes yet in your rights: Parliament is expressing a political position and asking the other institutions and the Member States to act. But it sets a course: if an administration uses AI to prepare a decision that concerns you, a person must be able to take back control, and AI must not decide alone in areas with legal or social effects. Calls for new rights (against deepfakes, for algorithmic transparency) and for better protection of elections would require revisions of treaties or laws, which are not decided by this vote. Parliament also calls for the European AI Act to be applied systematically and transparently.
The European Parliament has adopted a resolution requiring that a human genuinely supervise every AI tool used in the institutions and that any public decision assisted by AI can be taken over by a person. The text creates no binding rule: it is a political position, adopted by 405 votes to 133 with 60 abstentions, which sets a course but leaves the real questions of responsibility and cost untouched.
Parliament is right on substance: silently transferring democratic mandates to automated systems would be a serious drift. But a resolution binds no one. The principle of the human decision-maker, forcefully reaffirmed, does not say who controls, who answers for a bad decision, or how a harmed citizen obtains redress. Without a liability mechanism, the principle remains a declaration of intent.
The text asks that AI providers and deployers bear responsibility proportionate to their role, and that the burden not fall on end users. That is the right instinct: whoever decides must bear the cost. But as long as this allocation is not written into a binding act, it protects no one. The Commission is invited to apply the AI regulation systematically; it is not compelled to do so other than by political pressure.
The request to revise the Charter of Fundamental Rights in order to add AI-related rights is the most questionable point. Adding rights without defining who holds them, who owes them and what sanctions back them creates expectation without guarantee. A subsidiary approach would have consisted of checking whether existing rights (data protection, privacy, non-discrimination) are sufficient, and of strengthening their enforcement before opening a treaty revision process.
Parliament also wants a right to meaningful human intervention in AI-assisted administrative decisions. The intention is laudable, but the cost of this guarantee must be weighed: for an administration, it means procedures, delays and staff. The text does not say who pays, nor how to prevent this right from becoming one more queue. Simplifying existing avenues of appeal would have been more effective than proclaiming a new right.
The text insists on strategic autonomy and a preference for AI tools developed and hosted in the Union. That is a legitimate instinct of sovereignty, but it must not become a protectionist reflex. Technological sovereignty is built through a less fragmented internal market for AI, investment and open competition, not through a bureaucratic preference that would decide in place of buyers. Parliament says this in part, but the balance remains fragile.
The vote (405 for, 133 against, 60 abstentions) shows broad support for this orientation, without unanimity. The groups that voted against or abstained are not necessarily wrong on everything: the question of the administrative burden and the cost of new layers of control remains open. Parliament should have conditioned its demands for new rights and new control mechanisms on an assessment of their cost for administrations and businesses, and on the removal of redundant obligations.
A principle without sanction is not a guarantee, it is a wish.
The European Parliament has adopted a resolution requiring that a human genuinely supervise every AI tool used in the institutions and that any public decision assisted by AI can be taken over by a person. The text creates no binding rule: it is a political position, adopted by 405 votes to 133 with 60 abstentions, which sets a course but leaves the real questions of responsibility and cost untouched.
Parliament is right on substance: silently transferring democratic mandates to automated systems would be a serious drift. But a resolution binds no one. The principle of the human decision-maker, forcefully reaffirmed, does not say who controls, who answers for a bad decision, or how a harmed citizen obtains redress. Without a liability mechanism, the principle remains a declaration of intent.
The text asks that AI providers and deployers bear responsibility proportionate to their role, and that the burden not fall on end users. That is the right instinct: whoever decides must bear the cost. But as long as this allocation is not written into a binding act, it protects no one. The Commission is invited to apply the AI regulation systematically; it is not compelled to do so other than by political pressure.
The request to revise the Charter of Fundamental Rights in order to add AI-related rights is the most questionable point. Adding rights without defining who holds them, who owes them and what sanctions back them creates expectation without guarantee. A subsidiary approach would have consisted of checking whether existing rights (data protection, privacy, non-discrimination) are sufficient, and of strengthening their enforcement before opening a treaty revision process.
Parliament also wants a right to meaningful human intervention in AI-assisted administrative decisions. The intention is laudable, but the cost of this guarantee must be weighed: for an administration, it means procedures, delays and staff. The text does not say who pays, nor how to prevent this right from becoming one more queue. Simplifying existing avenues of appeal would have been more effective than proclaiming a new right.
The text insists on strategic autonomy and a preference for AI tools developed and hosted in the Union. That is a legitimate instinct of sovereignty, but it must not become a protectionist reflex. Technological sovereignty is built through a less fragmented internal market for AI, investment and open competition, not through a bureaucratic preference that would decide in place of buyers. Parliament says this in part, but the balance remains fragile.
The vote (405 for, 133 against, 60 abstentions) shows broad support for this orientation, without unanimity. The groups that voted against or abstained are not necessarily wrong on everything: the question of the administrative burden and the cost of new layers of control remains open. Parliament should have conditioned its demands for new rights and new control mechanisms on an assessment of their cost for administrations and businesses, and on the removal of redundant obligations.
A principle without sanction is not a guarantee, it is a wish.