In short

From 30 September 2026, Legislative Decree 160/2026 brings two AI-related offences into the Modello 231: failure to secure high-risk systems (art. 437-bis of the Criminal Code) and the unlawful spread of deepfakes (art. 612-quater of the Criminal Code). For most SMEs the first does not apply, the second does. First, find out who uses which AI, with what permissions, and who checks the output.

Key points

  • The new art. 25-vicies of Legislative Decree 231/2001 makes the company liable, not only the individual, for two AI-related offences.
  • Art. 437-bis covers only high-risk AI systems and applies only if the omission creates a concrete danger to life or physical safety. It is not the ChatGPT your sales team uses.
  • The deepfake offence (612-quater) can affect any company that produces content with AI.
  • "We do not process personal data" is not a defence: these offences do not depend on personal data, and even in B2B there is personal data anyway.
  • A company's defence in a 231 proceeding is to show it has rules: who uses what, with what permissions, and who checks.

From 30 September 2026, artificial intelligence enters the Modello 231, the organisational model under Legislative Decree 231/2001 on companies’ liability for offences. Legislative Decree 160/2026, published in the Official Gazette on 15 September, adds a new art. 25-vicies to Legislative Decree 231/2001: “Offences committed with the use of artificial intelligence systems”.

In plain terms: for two AI-related offences, if the act is done in the company’s interest or to its advantage, the individual is not the only one who pays. The company pays too.

Before anyone panics, though, read what the decree actually says. For most SMEs the real risk is not the one in the headlines.

What the decree says in brief

Two offences enter the 231 catalogue.

Art. 437-bis of the Criminal Code - failure to adopt security measures in artificial intelligence systems. It punishes whoever omits the technical security or human oversight measures required for high-risk AI systems, when the omission creates a concrete danger to the life or physical safety of people. For the company: a fine of 600 to 1,000 quotas (the units in which 231 corporate fines are calculated), plus disqualification penalties.

Art. 612-quater of the Criminal Code - unlawful spread of content generated or altered with AI. This is the deepfake offence, already introduced by Law 132/2025 (Italy’s national AI law). It punishes whoever spreads false AI-generated content (images, video, voices) without consent, causing unjust harm. From 30 September it involves the company as well.

What does not change for most SMEs

Art. 437-bis has two precise conditions: a high-risk system under the AI Act and a concrete danger to life or physical safety.

The ChatGPT your salesperson uses to write quotes is not a high-risk system. Neither is the Copilot that summarises meetings.

So it makes no sense to say “from tomorrow every SME that uses AI faces criminal charges”. It is not true, and whoever tells you so is selling fear.

But that does not mean the decree is not about you.

Why “it does not concern us” is still a mistake

The most common objection we hear is: “We are B2B, we do not process personal data, so this does not touch us.” It is wrong twice.

First: these offences do not depend on personal data. Art. 437-bis is about the safety of systems, art. 612-quater about false content. The GDPR is a different matter.

Second: even in B2B there is personal data anyway. The name and email of your client’s buyer are personal data. So are the CVs HR uploads to a tool for a first screening, the job reports with technicians’ names, and employee evaluations.

And here is the point that really matters. Some apparently ordinary uses of AI fall among the high-risk systems of the AI Act: recruitment, assessing workers’ performance, decisions on promotions or dismissals. If HR uses an AI tool to filter candidates, the question “are we high-risk?” is no longer theoretical.

As for deepfakes, they concern anyone who produces content with AI. A promotional video with the face of a client or a former colleague, generated without consent, is enough.

What those who govern should check: the 3 things to know by month end

You do not need a six-month project. You need to know three things that most owners cannot answer today.

1. Who uses AI, and with what permissions

The first question is not “which tool do we use”. It is “who can use what”. Who can upload internal documents, who can use AI on client and employee data, who can create agents that act on their own.

If the answer is “everyone can use it”, you have no openness. You have a company where nobody answers for anything.

2. With which tools, and whether any is high-risk

List the AI tools in use. The approved ones and the ones the team uses on its own. Then look at the list with a single question: does any of them decide or influence something about people (candidates, employees) or physical safety (plants, machines, logistics)?

If so, that is the system to bring under control first, with declared human oversight and a name next to it.

3. Who checks, before the output leaves the company

Any content generated with AI that leaves the company must pass through a person who checks it. This holds for texts, and even more for images, video and voices. It is the simplest rule and the one that protects most against art. 612-quater.

We put the minimum rules in a checklist: AI policy for SMEs: an operational checklist to avoid chaos.

If you already have a Modello 231

The Modello needs updating. The new art. 25-vicies is one more predicate offence, and the Supervisory Body (Organismo di Vigilanza, OdV) must assess whether your company faces a concrete risk and which safeguards cover it.

A company’s defence in a 231 proceeding is always the same: show that it had an adequate organisation before the fact. On AI, this means a written map of who uses what, with which permissions, and who checks. Without that map, the OdV has nothing to supervise.

If you do not have a Modello 231

The decree does not oblige you to adopt one. But the three questions above still apply, because the AI Act asks anyone who uses AI systems to govern them anyway. The full picture is in EU AI Act and Italian SMEs: what GMs and Fractional Managers need to know.

In short

  • Art. 437-bis is an offence for those who mismanage high-risk AI systems with a concrete danger to people. It is not your case if you use AI only to write and summarise.
  • Art. 612-quater, the deepfake, can affect anyone who produces content.
  • In both cases the defence is the same: know who uses AI, with which tools, with which permissions, and who checks.

To map who uses what, start from the register of AI tools. To decide where to put the controls, see where to put the human when AI works for you.

If you cannot answer these questions in 10 minutes, the problem is not the law. It is that nobody is governing AI in your company.

The quickest way to find out is the SME Diagnostic: in a few minutes you see where AI has already entered your processes and where a control is missing.

Primary sources and references

Frequently asked questions

What does Legislative Decree 160/2026 provide for SMEs that use AI?

It adds two offences to the 231 catalogue (new art. 25-vicies): failure to adopt security measures on high-risk AI systems (art. 437-bis of the Criminal Code) and the unlawful spread of content generated or altered with AI, meaning deepfakes (art. 612-quater of the Criminal Code). If the offence is committed in the company's interest or to its advantage, the company is liable too.

What must a company do by 30 September 2026?

Three things. Map which AI tools it uses and who uses them. Check whether any of them count as high-risk systems under the AI Act. Decide who checks AI-generated content before it leaves the company. A company that already has a Modello 231 should ask its Supervisory Body (OdV) to update it.

How do I find out which AI tools my team already uses?

Ask the team, without punishment, and check invoices and the AI features already active in the software you pay for. A one-page register of who uses what, and with which data, is the starting point.

Who is responsible if a high-risk AI system has no human oversight?

The person who omits the security or human oversight measures is criminally liable if this creates a concrete danger to life or physical safety. Under the new art. 25-vicies the company can be liable too, with a fine of 600 to 1,000 quotas (the units used to calculate 231 corporate fines) and disqualification penalties.

Sources

  1. Official Gazette - Legislative Decree of 9 September 2026, no. 160, art. 12 (new art. 437-bis of the Criminal Code)
  2. Giurisprudenza Penale - Legislative Decree 160/2026 published in the Official Gazette, in force from 30 September
  3. AI Act Service Desk - Art. 26, obligations of deployers of high-risk systems

Last updated:

Related articles

AI Act delay: only the high-risk rules slipped. Here is what still applies to an SME using AI today The Digital Omnibus delayed only high-risk AI rules. See what still applies from 2 August 2026 to an SME using AI, and why a register is not governance. AI policy for SMEs: the 7 controls to write, with examples on data, human review, roles and metrics The 7 controls of an AI policy for SMEs: allowed data, human review, responsibility, tracking, bans, metrics and processes. Before you sign an AI proposal: the questions to ask, from the side of whoever pays Got an AI proposal and nothing to compare it with? Ask about a similar case, size, how to explain the spend, and what stays after the contract.

Next Step

Reading is not enough if the system stays the same.

If this article touches a real source of friction, the useful next step is not consuming more content. It is understanding where you are losing time, control or margin today.

Take the SME Diagnostic → See the systems for companies →