This article has been originally published on ai4business.it on January 15, 2026 (italian only).

Artificial intelligence is a key tool for industry and business: it improves efficiency, safety, and sustainability, but poses challenges of reliability, transparency, and risk. Between black boxes, data validation, and governance, the AI Act and risk analysis become central for a credible, responsible, and truly advantageous adoption in complex industrial contexts and strategic corporate decisions.

Artificial intelligence is now a tool that, when used appropriately, opens up various opportunities from multiple perspectives, some even exceptional. For instance, it was recently used to solve a complex unsolved theoretical equation known as the Erdős number problem and to define an experiment aimed at verifying mutations in human immune cells.

Opportunities, data and industrial applications

Therefore, AI offers a wide possibility of use in different contexts and for different objectives, thanks to the abundance of data and information available today that can be processed with AI to extract knowledge to be used in marketing, as in finance and especially in industry, a specific sector of great relevance and in which the impact of AI can truly make the difference, even in terms of sustainability.

Industry can benefit from the processing of data and information to make processes more efficient, systems and components more reliable, and consequently, guarantee greater safety in industrial functioning and operations.

Moreover, efficiency, reliability and safety are elements of sustainability in the use of resources and contribute to avoiding accidents even with potential catastrophic consequences.

AI in particular contributes to safety by allowing the “reading” of data and signals from sensors, seeing images of thermographies, tomographies and photographs, to extract information on the state of components, monitor the system situation and be able to control it in real time, anticipating failures, resolving functioning anomalies in various industrial contexts.

AI also allows for the acceleration of simulations of accidental scenarios and the management in real time of emergency situations that can occur. All this facilitates a better overall management of safety.

The challenges of AI: credibility of the “black box”

Although the potential of AI in industry is almost obvious, its adoption in the field is not simple. AI requires data and information that, while being available, must be verified and validated to then be processed appropriately. The advancements and algorithms in this field are exceptional and for the large part exist, but they present a series of limitations that must be addressed.

For a practical use of the AI tool, it is essential to be able to trust the outputs of the algorithms. These algorithms are born as “black boxes”: it is not known how the model internally processes the input data to provide the output data which are then used to guide decisions. This opacity of the model worries in industrial situations, critical both for production efficiency and from the point of view of safety.

For example, if an AI algorithm, reading process parameters and sensor measurements on machines, suggests turning off the plant due to an imminent failure of a critical component, the operator must be certain of being able to take the responsibility of turning it off. It is important that the shutdown is effectively necessary, to avoid interrupting production and service unnecessarily. Therefore, it is crucial to be sure and confident in the response and in the suggestion of the algorithm, analogously to when one trusts the advice of a doctor after the reading of physical exams.

The credibility of outputs

Today, the primary objective in the industrial and business context is to reach the credibility of the outputs of artificial intelligence algorithms. The credibility of the results is indispensable so that they can be adopted in decisions on how to design, operate and maintain systems efficiently, and productions in a reliable and safe manner.

To reach this goal, research in AI is making great efforts to allow:

• to verify and validate the data used for the training of the algorithms.
• To guarantee the accuracy of the prediction and the output of the algorithm.
• To explain why the algorithm provides a specific indication on the state of the components and of the process.

Governance and risk analysis: the AI Act

For its importance, for its potential, the artificial intelligence tool must be governed in its development and in its use in a just way. In this sense, the AI Act must be seen as a measure that favors the right use of AI, without limiting it. Therefore, I repeat, the AI Act is an important element of governance of the development and use of AI, since it lays the foundations to operate and use the tool in a right manner.

In my personal opinion, the AI Act has a very important role in itself, but its transposition and employment must be equally relevant in practice. What still needs to be developed is an executive approach in support of the AI Act, specifically in terms of risk analysis. AI, in fact, while bringing great benefits, also entails risks related to its use.

The methodological approach of risk analysis must support the appropriate adoption of the AI Act. Risk analysis methodologies already developed and used for critical plants and systems that provide services and products to society already exist, and from these one can borrow.

How to approach risk analysis

Risk analysis must be addressed application by application and problem by problem. Qualitative, semi-qualitative and possibly quantitative tools are necessary to examine the impact of AI in the specific context, beyond the “risk” categories included in the AI Act.

It is necessary to:

• Identify the different types of stakeholders who can benefit or suffer the impact of the AI tool and define their characteristics.
• Understand to which dangers these well-categorized groups are exposed, and how much, due to the use of AI, its possible biases or its possible errors.
• Determine the level of exposure of the different groups potentially exposed (both for benefit and impact), considering their vulnerability and the probability of being subject to the consequences of the danger.

The adoption of the AI Act, with adequate evaluation structures, must guarantee that, on the basis of the knowledge of the exposure to dangers, the levels of exposure and the vulnerabilities of the elements that can be touched by the use of AI, such use is limited or adopted in the right manner.

An effort in this direction is fundamental so that AI truly brings great opportunities and benefits in all industrial and business sectors.