Since 2 August 2026 the AI Act (Regulation EU 2024/1689) is fully applicable. For Italian companies the operational phase begins: no longer preparation, but the duty to comply with deadlines and rules that had no effect until now.
What Changed on 2 August 2026
The rules divide AI systems into four categories, and the deadlines have unfolded in sequence:
- Unacceptable risk: Prohibited (e.g. social scoring), applicable since 2 February 2025. The bans on non-consensual imagery and child sexual abuse material apply from 2 December 2026.
- High risk: Strict obligations (risk assessment, technical documentation, human oversight, post-market monitoring). Annex III high-risk systems apply from 2 December 2027; Annex I systems, i.e. regulated products requiring certification, from 2 August 2028.
- Limited risk: Transparency obligations. Since 2 August 2026 Article 50 applies: inform users they are talking to a machine, mark synthetic content, identify deepfakes.
- Minimal risk: No particular restriction, but the AI literacy obligation in the workplace has been in force since 2 February 2025.
The genuinely new point: transparency
The concrete novelty of August 2026 does not lie in the prohibitions, operative for some time, but in the transparency obligations of Article 50. In practice, every chatbot addressed to customers or consumers in Italy must make it clear that it is not speaking with a person, and AI-generated or manipulated content must be marked as such.
Operating this way without complying does not merely risk a sanction: it exposes the company to competition-law challenges, since statements made to the user through an automated system remain attributable to the service provider.
General-purpose models and the AI Office
Since 2 August 2026 the AI Office may exercise its enforcement powers over general-purpose AI models (Articles 88-90). Models placed on the market before 2 August 2025 have until 2 August 2027 to comply fully with the Regulation; the Code of Practice for general-purpose AI models was declared adequate on 1 August 2025.
Member State obligations
By 2 August 2026 every Member State had to establish at least one national regulatory sandbox (Article 57), and systems already marketed before 2 August 2026 must comply with Article 50(2) by 2 December 2026. Italy has kept pace on this front.
Penalties and outlook
Violations can carry fines of up to EUR 35 million or 7% of total annual global turnover, depending on the risk category. For general-purpose models the law even provides, in extreme cases, for the withdrawal of the model from the market.
The operational step now is no longer reading the Regulation but answering three questions: which AI systems do we actually use, and in which processes do they affect decisions concerning people; how do we tell users they are interacting with an automated system; and who is accountable for compliance. For most Italian companies that mapping has never been done.
Is your company compliant with the AI Act?
With full application since 2 August 2026, compliance is no longer a project: it is a standing obligation. We provide legal audits of your systems and documentation.
Request an Audit