Mentre USA e Cina corrono senza freni sull'IA, l'Europa accumula ritardi: Gates chiede un patto globale urgente

La corsa all’intelligenza artificiale tra Stati Uniti e Cina allarga il divario sulla sicurezza, mentre in Europa l’AI Act resta in gran parte sulla carta. Bill Gates invoca regole federali e un patto globale: quali controlli comuni possono arginare i rischi?

Bill Gates told NBC’s Meet the Press on September 27, 2026 that artificial intelligence is powerful enough to trigger events causing a billion deaths. The Microsoft co?founder warned that self?regulation by tech companies will not stop malicious actors from exploiting advanced models, and he urged the United States to adopt a federal AI law with built?in guardrails.

Why Gates pushes for US federal AI regulation

Gates framed the threat as a matter of national security, arguing that only legislation backed by law enforcement can enforce safety standards. He called self?regulation insufficient and proposed a federal framework that would require:

• certification of high?risk models
• mandatory transparency reports
• and clear liability rules

The interview highlighted a growing political divide: while Congress has yet to pass comprehensive AI legislation, several states are drafting their own rules, creating a patchwork that could hinder innovation and leave gaps in protection. Gates emphasized that regulation should not stifle progress but add safety barriers that keep pace with rapid technological advances.

US?China race widens the safety gap

The competition between Washington and Beijing is accelerating AI development at a pace that outstrips regulatory efforts. Both nations prioritize speed and strategic advantage, resulting in minimal oversight of powerful models. At the United Nations in September 2026, OpenAI’s Sam Altman and Anthropic’s Dario Amodei urged the Security Council to adopt global testing standards and an incident?notification system, yet the United States rejected any binding governance, insisting that a multilateral body would limit its sovereignty. China echoed similar concerns, opposing what it called “globalist control” of its technology sector. Without mandatory reporting or shared safety benchmarks, dangerous capabilities can spread unnoticed, increasing the risk of misuse in:

• cyber?attacks
• biological research
• or disinformation campaigns

Europe’s AI Act is on paper but not on the ground

The European Union finalized the AI Act in 2024, with key provisions becoming applicable in 2025 and the full regime scheduled for 2026. However, enforcement capacity remains limited: many member states lack dedicated AI oversight bodies, and companies are still adapting to the new classification system. Investment data from OECD shows European AI funding trails both the United States and China, reflecting a broader lag in research and deployment. While the Act introduces risk?based rules and mandates for high?impact systems, the delay in practical implementation means that European firms often operate under looser interim guidelines, reducing the continent’s ability to influence global standards.

What common controls could address the risks

A workable safety framework would combine independent pre-release testing, transparency about training data and known limitations, mandatory incident reporting, capability thresholds for the most powerful models, and clear liability rules. International coordination could make these controls more effective, but political resistance has so far prevented formal agreement. The immediate priority is to establish enforceable domestic standards and shared procedures for reporting serious incidents.

Leggi anche: Bill Gates avverte sui rischi dell'intelligenza artificiale non regolamentata