Italy advances EU AI Act implementation with new national AI decree

Italy is moving quickly to turn the EU AI Act from Brussels‑level text into concrete national rules, and this decree shows how that will feel at street level. The government is drawing bright lines in three politically sensitive areas: policing, employment, and system safety. For police, AI becomes a decision‑support tool, not an autonomous actor—human oversight is mandatory and mass biometric surveillance is explicitly off the table. For employers, core employment decisions such as hiring, contract changes and dismissal cannot be left solely to algorithms, and any firing that relies exclusively on AI is declared null.([finanza.repubblica.it](https://finanza.repubblica.it/News/2026/06/10/decreto_ai_via_libera_preliminare_del_cdm_focus_su_sicurezza_lavoro_e_giustizia-167/))

The introduction of a new criminal offence for those who design, deploy, or fail to secure AI systems in ways that create concrete risks to people or the state is particularly notable. It personalizes accountability for negligence in AI safety, pushing responsibility up the stack from mere corporate fines to potential criminal liability. That’s likely to make boards and product leaders far more conservative about high‑risk deployments in Italy.

For the race to AGI, this doesn’t slow core model research so much as it increases the friction around certain classes of deployment, especially in law enforcement and HR. But because Italy is framing itself as the first country with an “organic” national AI law layered on top of the EU framework, its approach could become a reference point for other member states. Over time, that may influence how and where frontier‑level systems are productized in Europe versus looser jurisdictions.