AI in the Company: Internal Policies Lagging and What the Italian Framework Requires
Generative AI moves faster than company policies. What Decree 179/2026 requires on automated decisions, transparency, monitoring and shadow AI.

The use of AI in companies is growing faster than the internal policies designed to govern it. For Italian businesses the reference is Legislative Decree No. 179 of 7 October 2026, which implements the AI Act, the European regulation on artificial intelligence.12 The starting point indicated is visibility: contracts to update, unauthorised tools to identify, and training that is not merely cosmetic.1
What shadow AI is and why it is a concern
Shadow AI is the use of artificial intelligence tools that the company has neither registered nor authorised. According to AI4Business, it can lead to company and personal data being sent to third-party servers, prompts containing confidential information or trade secrets, results used without verification, and tools whose use triggers legal obligations.1 AI4Business gives the example of a team manager who entrusts an AI assistant with data on employees’ access and working hours, in order to receive reports on anomalies or drops in productivity. In practice, this would activate remote monitoring without asking whether a union agreement or authorisation from the Labour Inspectorate is needed.1
What Decree 179/2026 provides for employment relationships
Final decisions on the establishment, modification or termination of an employment relationship, including disciplinary measures and performance evaluations, cannot be based solely on automated processing. The decision remains with a natural person who exercises effective and independent power.1 The worker’s consent does not make an exception to this rule. Dismissal or other employer decisions taken in breach of these principles are void.1 The express exception concerns the search and selection of candidates, even when it results in a candidate not being admitted to the subsequent stages.1 On the safety front, the decree brings artificial intelligence into the risk assessment under the Consolidated Workplace Safety Act whenever a system affects work organisation, production rhythms or the way work is performed. In these cases, an obligation to provide specific information and training on the related risks is also foreseen.1
Transparency and remote monitoring: what workers can ask for
If an automated decision or monitoring system provides relevant indications on hiring, management, job assignments, supervision or evaluation, the employee has the right to know that it exists, how it works and which data it uses. Trade union representatives have the right to the same information in a structured form.1 If AI weighs on a specific decision, the worker may ask, case by case and through a natural person, for a comprehensible explanation. They must be told to what extent AI systems influenced the decision and what the main parameters were.1 The constraint of Article 4 of the Workers’ Statute also remains for tools that allow, even only indirectly, remote monitoring. An agreement with the trade unions is required or, if none is reached, the approval of the Labour Inspectorate. Tools used to perform the work and those that record access and attendance are excluded.1 With generative AI integrated into increasingly complex business software, this boundary is not always easy to draw. The analysis must be carried out case by case before a feature is activated, not afterwards.1
Why banning generative AI is not enough
According to AI4Business, a general ban fuels shadow AI: if the employee finds the tool useful and the ban a bureaucratic obstacle, they will use it anyway, perhaps from a personal device and without supervision.1 An effective policy indicates which tools are authorised, which information must not enter prompts and who checks the results before they are used.1 Lawyer Marco Panichi, quoted by the agency AGI, identifies a census as the first step. It is necessary to reconstruct which tools the company and its staff use, including those purchased as a service, with which data and from which suppliers. Internal rules are then defined on this basis.2
The numbers behind the lag
Panichi also cites the Lenovo-Confindustria Giovani research, presented in June: 82.4% of companies admit they do not have adequate skills in-house. According to Istat, among companies with at least ten employees, the share using AI rose from 8.2% in 2024 to 16.4% in 2025, roughly doubling.2 The deadline is approaching. On 2 December the new prohibitions set out in Article 5 of the AI Act take effect, and the period for marking AI systems already on the market ends. AGI notes that this is the first European deadline since the supervisory authorities became operational in Italy, and that they can impose fines of up to 35 million euros.2 Panichi clarifies that the maximum amounts are an upper limit and not a tariff. Before even considering the amount, a company must know which systems it uses and in what role, as provider or deployer.2
Training and employee trust
According to the HR Trends research by Randstad Professional, reported by Agenda Digitale, seven out of ten Italian workers fear, at least in part, becoming professionally obsolete. The article adds that AI has entered the workflows of more than half of companies, mostly in an unstructured way.3 For Agenda Digitale, without listening and transparency this uncertainty can turn into demotivation and hold back the expected benefits of investments.3 On the operational side, Emanuele Marcianò, CEO of DUNE, interviewed by ZeroUno, identifies sharing of tools, management leadership and IT support as factors favouring adoption. In his view, strategic choices remain with people.4
What changes
Training appears both as an obligation linked to workplace safety and as a factor of employee trust.13
This article was written by artificial intelligence from the sources cited, without human review. Automatic newsroom by Redazia.
Sources
Footnotes
-
AI e lavoro: la governance che serve alle aziende, AI4Business, 2026-11-17 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 ↩11 ↩12 ↩13 ↩14 ↩15
-
AI Act in Italia, a che punto sono le imprese, agi.it, 2026-11-18 ↩ ↩2 ↩3 ↩4 ↩5
-
La paura di diventare obsoleti può frenare l’AI in azienda: cosa fare, Agenda Digitale, 2026-11-18 ↩ ↩2 ↩3
-
AI nei processi: partire dal ritorno sugli investimenti, ZeroUno, 2026-11-18 ↩


