Illustrazione editoriale: un sistema di IA capace di agire in autonomia entro il perimetro delle regole tecniche e giuridiche

Immagine creata con IAAI-generated image

Artificial Intelligence systems no longer merely generate content: they act, they send, they modify, they commit. This is the starting point of “Agentic AI: Technical Autonomy, Human Responsibility, and Legal Governance”, the book by Nicola Fabiano published in July 2026, whose thesis is declared from the opening pages and runs through the entire work: technical autonomy does not create legal autonomy.

Technical standardisation is the bridge between the requirements of the AI Act and their verifiability. Without harmonised standards, the presumption of conformity remains a promise on paper.

Valentina Grazia Sapuppo
Law & Technology

The book

The work is organised into eighteen chapters across four parts, devoted respectively to the foundations, to risks and legal qualification, to responsibilities and evidence, and to the operational governance model, followed by the Conclusions and by an apparatus designed for professional use: the reusable scheme of the seven column matrix (Appendix A), two original Python tools reproduced in full and in working order (Appendix B), a reasoned glossary, the table of legal sources and the analytical index. The Italian paperback edition runs to 310 pages (ISBN 979-12-243-4018-8), alongside the digital EPUB edition (ISBN 979-12-243-4019-5). The book is also published in English (ISBN 979-12-243-4020-1 for the paperback edition). The preface is by Antonino Caffo. The work is accompanied by the open source repository “agentic-control-layer”, released under the EUPL-1.2 licence, which is its reference implementation.

Limits before, evidence during, responsibility after

The method proposed rests on three questions that every organisation should ask itself the moment it delegates to a system the capacity to act: which limits to set before the action, which evidence to retain during it, and who must be able to answer for it afterwards. Around these questions the book builds original categories, namely graduated agenticity, the chain of attribution of the agentic action and evidentiary accountability, and hands them over to an operational tool, the seven column matrix, taken as far as being executable in code. The merit of this construction, for anyone working on compliance, is that agenticity is treated as a scale and not as a label: governance is calibrated on the degree of autonomy actually delegated, not on the commercial name of the product.

Chapter 16: from technical requirements to the presumption of conformity

Chapter 16, “Technical standardisation for agentic systems: from technical requirements to the presumption of conformity”, is by the present writer. It starts from the mechanism of harmonised standards and the presumption of conformity under Article 40 of the AI Act, and then reconstructs the fora in which standardisation is produced, from ISO/IEC to CEN-CENELEC, and the Italian role in that work through UNINFO. It then surveys the technical documents that systems capable of acting call into question in a specific way, from event logging to human oversight, from transparency to risk management, and on to conformity assessment. This is the point at which the law stops stating principles and starts asking for proof: a terrain that, for the most part, is still under construction.

Press coverage

The book has been covered by Stato Quotidiano, on 9 July 2026, and by RAI News, which on 10 July 2026 published Pierluigi Mele’s interview with Antonino Caffo, entitled “AI agents and legal liability: who answers when the machine acts on its own?”. The book is also presented in issue 29 of the NicFab Newsletter, of 14 July 2026.

Conclusions

Delegating the capacity to act to a system does not shift responsibility: it merely makes it harder to reconstruct. An organisation that has not decided beforehand which actions to allow, and has not designed during the action which evidence to retain, will find afterwards that it can prove nothing. One wonders, then, whether undertakings and public administrations will be able to put this method into practice before incidents impose it on them, and whether technical standardisation will be able to keep pace with systems that, while we are still discussing how to qualify them, have already started to act.

Cover of the book Agentic AI by Nicola Fabiano
The book
Agentic AI

Technical Autonomy, Human Responsibility, and Legal Governance · Nicola Fabiano · preface by Antonino Caffo · paperback and EPUB · Italian and English editions

PaperbackDigital edition. Available on Amazon, Apple Books, Kobo and other stores.
Where to buythe full, up-to-date list of stores for both editions is on the official book page.

AI AnthropoCosmic In evidenzaAI AnthropoCosmicCall for Paper aperta fino al 15 settembre 2026. Un progetto internazionale per un’IA a servizio dell’Uomo, dell’Ambiente e del Cosmo. Leggi l’articoloAI Open Mind AI AnthropoCosmic FeaturedAI AnthropoCosmicCall for Paper open until 15 September 2026. An international project for an AI at the service of humanity, the environment and the cosmos. Read the articleAI Open Mind Agentic AI In evidenzaAgentic AILimiti prima dell’azione, evidenze durante, responsabilità dopo. Il volume di Nicola Fabiano sulla governance dei sistemi agentici, con la prefazione di Antonino Caffo.Capitolo 16 a cura dell’Avv. Valentina Grazia SapuppoLeggi l’articolo Agentic AI FeaturedAgentic AILimits before the action, evidence during, responsibility afterwards. Nicola Fabiano’s book on the governance of agentic systems, with a preface by Antonino Caffo.Chapter 16 by Valentina Grazia SapuppoRead the article