On 28 October 2026, Pasquale Stanzione, president of Italy's data protection authority, the Garante per la protezione dei dati personali, testified before the XIV Commission of the Chamber of Deputies on the Commission's proposed EU Kids Act, COM(2026) 681 final. The Garante's published account of the hearing shows a regulator endorsing the proposal's direction while insisting on a constraint that child-safety debates often skip: protecting children must not create a new database of children.
The strongest case for the proposal
The argument for the Kids Act is serious. Platforms optimised for engagement can exploit developing minds. Self-declared ages are trivially false. Chatbots designed to flatter and mirror a user can stand in for human relationships. If voluntary design changes had worked, regulators would not be legislating. Stanzione accepts much of this. He welcomed the proposal's measures on addictive design, harmful content exposure and unlimited autonomous interaction, and he described the three-tier model of differentiated age thresholds for social networks as more proportionate than a single hard cut-off, though complex.
Where the Garante draws the line
Stanzione's central condition is data minimisation. Age verification is acceptable, he said, but the data collected must be limited to what verification strictly requires. He objected to parental-supervision and monitoring tools that process personal data "eccedenti le effettive esigenze di verifica", meaning beyond what verification needs. He also asked that the European Data Protection Board and the EDPS be consulted on the delegated acts that will define which age-assurance methods are allowed. The proposal itself points toward privacy-preserving techniques such as cryptographic verification or zero-knowledge proofs, which confirm that a user is over a threshold without revealing identity.
This matters for innovation as well as civil liberties. If age assurance is built as document upload and face scanning stored by every platform, the result is a honeypot and a compliance moat that only large firms can afford. If it is built on minimal-disclosure proofs, small services can comply without becoming data brokers. A proportionate rule is also a pro-competition rule.
Chatbots that simulate relationships
The hearing's most novel passage concerned AI companions. Stanzione welcomed limits on chatbot features that simulate interpersonal relationships and create emotional dependence. He linked this to the "effetto Eliza" and warned of a "loop dell'empatia infinita": endless, flattering empathy that could make real human relationships feel unsatisfying and foster isolation.
The Garante has enforcement history here. In a decision dated 10 April 2025, it fined Luka Inc., the company behind the Replika chatbot, €5 million. The EDPB's summary records that the company had no age verification at registration or during use until 2 February 2023, despite saying minors were excluded, and that technical assessments found its current age-verification system still deficient. The Garante's press release of 19 May 2025 adds that it opened a separate investigation into the training of the generative model behind the service, and that the fine is under appeal.
The Replika case is a useful template because it was enforced under existing law. The GDPR already requires a legal basis, transparency and data protection by design. A regulator did not need a new chatbot statute to act on a service that exposed minors to a product with weak age gating. That is an argument for enforcing current rules before layering on new categories, and a caution against defining "simulated empathy" so loosely that ordinary assistants or customer-service tools are caught.
Bans are not enough
The closing theme was the most quotable. "Nessun divieto, infatti, potrà mai sostituire la necessità di una «paideia digitale»", Stanzione said: no prohibition can replace the need for a digital education built at school and at home. That is a notable position for a regulator, and it fits the evidence-based, speech-friendly approach. Access restrictions displace behaviour, can be circumvented, and do little to teach a teenager to recognise manipulative design. Education and age-appropriate defaults address the cause rather than the symptom.
The tension Stanzione did not resolve
The hearing sits awkwardly beside a parallel debate. Netzpolitik reported on 17 November 2026 that a European Trade Union Institute study of the Digital Omnibus finds the package weakens workers' data protection rights. The Commission presents the Omnibus as a simplification measure and estimates up to €5 billion in administrative savings by 2029, while more than 100 civil society organisations have called it an attempt to covertly dismantle protections. Brussels is simultaneously asking the GDPR to do more for children and less for everyone else. A regulator can credibly demand strict minimisation in age checks only if the underlying rules stay intact.
There is also a real implementation risk. The Kids Act leaves key details to delegated acts. If those acts permit intrusive verification, Stanzione's minimisation principle will have been endorsed in testimony and lost in drafting.
What to watch
- Delegated acts. Whether the EDPB and EDPS are formally consulted on approved age-assurance methods, as the Garante requested.
- Method neutrality. Whether privacy-preserving proofs are treated as sufficient, rather than a fallback behind ID upload.
- Definition of relationship simulation. Whether the chatbot provisions target design features that foster dependence, rather than conversational AI in general.
- Education funding. Whether member states resource the digital-literacy side that Stanzione says no ban can replace.
The Garante's testimony is useful because it keeps three goals in one frame: protecting minors, limiting data collection, and not mistaking prohibition for policy. Lawmakers should treat the minimisation test as a design requirement for every provision, not a footnote.