Artificial intelligence is delivering benefits that were unimaginable just a few years ago. Risks do exist, and they are not merely being raised by critics: they have recently been highlighted by the leaders of Big Tech themselves, and Geoffrey Hinton – the 2024 Nobel laureate in Physics and one of the pioneers of this technology – does not rule out catastrophic scenarios for humanity. But if we can manage these risks, AI can truly help us live in a better world: shortening the time it takes to discover drugs and vaccines, finding new treatments, refining diagnoses, flagging dangers, anticipating breakdowns and disasters, preventing landslides, and reducing water system leaks and traffic accidents.
And yet, at the same time, it is the most dangerously powerful tool ever placed in anyone’s hands. A paradox: the very technology that can nip an epidemic in the bud can also aid those who wish to turn a virus into a weapon. AI-based tools for synthetic biology do not distinguish between angels and demons: they have the same capacity to create both.
The recent report by the Lancet Commission on global health threats in the 21st century, among the many it lists between now and 2100, identifies two with the potential to wipe out humanity: nuclear war and the malicious use of AI.
But who is responsible for AI today? Those who regulate it, those who develop it, or those who use it? And who is accountable for its errors, or for the crimes that its use facilitates?
Who owns the Moon?
The question is less outlandish than it seems. The 1967 Outer Space Treaty establishes that space – including the Moon and other celestial bodies – is not subject to any national sovereignty: all states may explore and use it, but the Moon and other celestial bodies may be used only for peaceful and scientific purposes.
We need only look, for once, at the Moon and not the finger.
There is another precedent that speaks to the present day. When the first civilian airlines emerged in the early 20th century, the sky belonged to everyone and therefore to no one: a boundless space that closely resembles the digital realm. The issue of universal regulation arose immediately. As early as 1910, at the Paris Conference, nineteen nations began discussing the principles of overflight, international air law, and sovereignty over national airspace. It took decades, conventions, and a world war, but eventually, with the 1944 Chicago Convention, shared rules were established, with rights and responsibilities recognized by all.
And what about AI? From a legal standpoint, the starting point is clear: artificial intelligence is not accountable for its own decisions. Responsibility lies with the people and organizations that design, distribute, and use it.
There is also an aspect of generative AI that is as crucial as it is underestimated: the democratization of dangerous capabilities. Until recently, isolating a virus required highly specialized scientific training, years of laboratory work, sophisticated equipment, and knowledge accessible to only a few. Today, some of that knowledge can be reconstructed – or even suggested – by an advanced model. This does not mean that AI makes such an outcome easy or likely; it means that it lowers the informational barriers that had protected us until now. And this is a real risk.
Another concern is persuasiveness. Historian Yuval Noah Harari has been saying this for some time: a machine that masters language can build a relationship of intimacy and trust with people, and from there, shape their beliefs and behaviors. On the scientific front, Yoshua Bengio, a Turing Award laureate who chairs the International AI Safety Report, has warned about manipulative and deceptive behaviors already observed in the most advanced systems. This is exactly what chatbots are capable of doing, with the most vulnerable minds on the front lines. This is not a theoretical concern: numerous lawsuits have already been filed against Big Tech companies over this issue.
The issue, then, is not merely who commits the crime. It is who has the duty to prevent the circulation of tools capable of facilitating activities that are harmful to individuals or catastrophic for humanity. To whom should responsibility be attributed?
Not just ex post liability, but also preventive liability. Anyone who develops a very powerful model must be able to demonstrate that they have assessed and reasonably mitigated the risks arising from its capabilities, even before bringing it to market.
The European “AI Act” expressly recognizes this risk. For the most advanced general-purpose models—those classified as posing systemic risk – Article 55 requires providers to conduct assessments, adversarial testing, implement mitigation measures, and report serious incidents: obligations that will take effect in August 2025. Moreover, the European Commission no longer views large-scale generative AI as merely a technology that produces content, but as an infrastructure that organizes access to knowledge. The fundamental question remains: what controls and safeguards make the reliable adoption of AI possible?
We need globally agreed-upon rules, with certification bodies and regulatory agencies. We need standards that clearly define everyone’s responsibilities: those of the government, the manufacturer, the developer, and the user.
And we need greater awareness among those who use AI, from companies to individual citizens. Relying on models developed in Europe – which adhere to a regulatory framework built on the protection of individuals and data – is no longer a choice that can be put off.
* Valeria Sandei is Chief Global AI di Almaviva Group e President & CEO di Almawave Labs, a Group company active in the field of Data & Artificial Intelligence. Among her other roles, she serves as Chair of The Data Appeal Company, SisTer, Almawave do Brasil, Almawave Corp, and Mabrian Technologies. She is also a member of the Board of Directors of Tivit, a Brazilian IT company that joined the Group in 2025, and of Almaviva Solutions, a pillar of the Brazilian IT market.
Also in 2025, Sandei received the Italia Informa Award for Italian Excellence, the Key4biz Award for Artificial Intelligence, and the John McCarthy Award. She has been included in Fortune Italia’s list of the 100 most influential women of 2024 and in Forbes’ list of the 100 best managers of 2024.
She is also a member of the B7 Advisory Board, the Unicredit Centro Advisory Board, and the Astrid Observatory on Artificial Intelligence; she is a co-author of three U.S. patents; and she serves on the Scientific Committee of the AI & Business Strategy MBA program at the RCS Academy Business School






