The new security architecture

The evolution of Artificial Intelligence has redefined the architecture of global security. Today, the focus of defence is no longer only the protection of perimeter infrastructures, but the safeguarding of cognitive models, data integrity, and automated decision-making processes. It is a complex scenario where technological innovation must proceed hand in hand with the protection of rights, an imperative authoritatively recalled by the President of the Republic Sergio Mattarella in his recent speech at the Festival of Regions in Florence.

Innovation and protection of rights

As emphasised by the Head of State, governing artificial intelligence means enhancing its immense opportunities while preventing its uses from harming people. In this sense, European AI rules are not an obstacle but confirm a pioneering regulatory framework at the international level.

Shadow AI and human supervision

At corporate and governmental levels, this translates into the urgent need to contain the uncontrolled use of unauthorised applications (commonly known as “Shadow AI”). The use of algorithmic systems today requires verifiable governance and documented human supervision mechanisms. We must ensure that innovation serves to increase collective well-being and does not translate, as the President warns, into forms of dominance concentrated in the hands of a few. Breaches of security and transparency measures expose organisations today to severe sanction and reputational risks.

The training supply chain and threats

From a strictly cyber perspective, the attack perimeter has extended to the algorithm training supply chain. Insidious techniques such as Data Poisoning and Model Inversion attacks aim to secretly alter data sets to deceive the AI in production. This technical vulnerability finds a dramatic expression in the President’s alarm: the manipulation of opinions through fake news.

Fake news and information integrity

The malicious use of AI to generate manipulated information and false news not only influences commercial behaviours but risks confusing people’s judgement and public opinion, threatening the quality of information. Data security, in this context, transcends technical value to become a fundamental democratic guarantee.

AI agents and operational autonomy

The real qualitative leap in threats in 2026 is represented by AI agents endowed with operational autonomy. These entities have increasing privileges, with write access to code repositories and cloud systems. This delegation, if without rigorous controls, exposes to systemic incidents.

Secure by Design and Zero Trust

Accepting the institutional warning not to abdicate responsibility, it is imperative that infrastructures evolve towards a “Secure by Design” paradigm. It is essential to extend the principles of Zero Trust architecture to non-human identities, implement semantic checks inline before output execution, and guarantee full visibility of machine-to-machine interactions.

Preserving the cognitive infrastructure

Today’s cybersecurity challenge is not limited to repelling cyber intrusions but consists of preserving the reliability of our cognitive infrastructure. A mission that fully responds to the call of the institutions: to guide innovation while keeping security, fairness, and the fundamental rights of the person at the centre.

Sources and documents

Document date or reference period: Original contribution provided to the editorial team · 26 October 2026.

Signed contribution by Gerardo Costabile, provided to the editorial team. Author’s text published with only editorial layout; analysis and assessments belong to the author.

Report an update or correction →