Criminal Law Observatory · – 23 September 2026
From 30 September 2026, failure to adopt either the technical security or the human oversight measures for high-risk AI systems, where danger results, is a criminal offence. So provides the d.lgs. (legislative decree) 9 September 2026, no. 160, in the Gazzetta Ufficiale, Serie generale, no. 214 of 15 September 2026. Criminal exposure shifts onto system governance; the offence enters the 231 catalogue of reati presupposto (predicate offences).
The scope. The decree implements the delegation under art. 24, commi (subsections) 1, 2, letter h), 3 and 5, of the law of 23 September 2025, no. 132, adapting to regulation (EU) 2024/1689. It is neither the AI Act nor the national AI statute, but the implementing segment on policing (Title I) and criminal and civil liability (Title II).
Art. 12 introduces into the codice penale (Italian criminal code) art. 437-bis, «Failure to adopt security measures in artificial intelligence systems and unlawful alteration of the systems». Comma 1 punishes by imprisonment from one to five years the failure to adopt the technical security measures required for designing, training, producing and placing high-risk systems on the market (measures apt to prevent malfunctions or alterations in their operation), or, alternatively, the human oversight measures, where either omission creates danger to life or to public or individual safety; from two to eight years where the danger concerns State security. Comma 2 punishes the alteration of high-risk systems giving rise to the same danger, unless the act is a more serious offence and outside the cases of comma 1: imprisonment from two to six years, from three to ten years where it concerns State security. Comma 3 makes only the comma 1 case punishable for gross negligence, the penalty reduced from one third to one sixth; commi 2 and 4 require intent. Comma 4 extends the comma 1 penalties to professional users who intentionally omit human oversight.
Art. 15 inserts art. 25-vicies into the d.lgs. 8 June 2001, no. 231, after art. 25-undevicies: from six hundred to one thousand quote (penalty units) for art. 437-bis c.p., from two hundred to seven hundred for art. 612-quater c.p., with the disqualifying sanctions under art. 9, comma 2, letters b), c), d) and e).
Why it matters. The decree introduces neither art. 612-quater c.p. nor the aggravating circumstance under art. 61, no. 11-decies, c.p., both from art. 26 of the law no. 132/2025; art. 612-quater becomes a reato presupposto. Danger, an element of the offence in every case, must be proved in its causal derivation; the «high-risk» classification is the first ground of defence.
In practice. No deadlines bind undertakings: 30 September 2026 is entry into force, not a due date.
Administrative body: resolution mapping the AI systems in use and classifying risk under art. 6 of the regulation.
Compliance function: modello organizzativo 231 (organisational model), special part extended to art. 25-vicies; protocols on human oversight and logging; supply clauses on technical measures and traceability.
Organismo di vigilanza (supervisory body): dedicated information flows.
Summary note by Studio Legale Petrali, based on the text of the decree as published in the Gazzetta Ufficiale. The text does not reproduce original contributions and does not constitute professional advice.