As artificial intelligence becomes increasingly embedded in business operations, companies face a growing challenge: ensuring that AI systems are properly supervised and that the effectiveness of their controls can be demonstrated. Against this backdrop, Italy has taken a further step in aligning its national legislation with Regulation (EU) 2024/1689 (the AI Act), introducing new AI-related offences into the corporate liability framework under Legislative Decree No. 231/2001.

New Offences: From AI Safety to Deepfakes

Legislative Decree No. 160 of 9 September 2026, which entered into force on 30 September 2026, introduced Article 25-vicies, entitled “Offences Committed Using Artificial Intelligence Systems”, extending corporate liability to two offences concerning AI safety and the unlawful dissemination of AI-generated or AI-altered content:

• Article 437-bis of the Italian Criminal Code criminalises certain failures to implement technical safety measures or human oversight mechanisms in connection with high-risk AI systems, as well as unlawful interference with their operation, where such conduct creates a concrete risk to life, individual or public safety, or national security. The provision also covers certain omissions resulting from gross negligence and establishes specific rules for professional users who intentionally fail to implement the required human oversight measures.
• Article 612-quater of the Italian Criminal Code addresses the phenomenon of deepfakes by criminalising the non-consensual dissemination of AI-generated or AI-altered images, videos or voice recordings that are capable of misleading others as to their authenticity and cause unjust harm to the person concerned. Unlike the first offence, which specifically concerns high-risk AI systems, the deepfake offence is not subject to the same limitation and may therefore also be relevant to ordinary business activities, including corporate communications, marketing and digital content creation.

What Is at Stake for Companies?

The new Article 25-vicies introduces significant monetary penalties for companies: from 600 to 1,000 quota units for offences under Article 437-bis and from 200 to 700 quota units for offences under Article 612-quater, calculated under the penalty system established by Legislative Decree No. 231/2001. Both offences may also result in disqualification measures, including restrictions on contracting with public authorities, exclusion from public grants and financing, and prohibitions on advertising goods or services. As with other offences under Legislative Decree No. 231/2001, corporate liability arises only where the relevant statutory requirements are met, including that the offence is committed in the interest or to the advantage of the company.

Turning AI Risks into Effective Compliance

The introduction of these offences brings AI-related risks into sharper focus for corporate compliance programmes. While updating the 231 Model is not required, companies should assess their exposure to the new risks and the adequacy of existing controls, starting with a mapping of the AI systems developed or used across business processes and their potential classification as high-risk under the AI Act. Depending on the AI systems involved, companies should focus on different areas of risk:

• Companies developing, producing or professionally using high-risk AI systems should focus on technical safety measures, human oversight, incident management and the allocation of responsibilities within the organisation.
• Companies using generative AI should also consider the risks associated with the creation and dissemination of AI-generated or AI-altered images, videos and audio content, ensuring adequate authorisation procedures, content approval mechanisms and oversight of external service providers.

Depending on the outcome of the risk assessment, companies may need to revise their internal procedures, strengthen existing controls and provide targeted training to the personnel involved.

From Oversight to Evidence: The Supervisory Body’s Role

The Supervisory Body (Organismo di Vigilanza) also plays a key role in monitoring compliance with the 231 Model. This may require structured information flows on AI-related risks, incidents and monitoring activities, supported by the relevant technical functions. Ultimately, effective AI compliance is not just about having controls in place, but about being able to demonstrate that they work. Clear responsibilities, documented oversight and traceable decisions are therefore essential to demonstrating that the safeguards established under the 231 Model are effectively implemented.