A new study published in a reputable scientific journal Proceedings of the National Academy of Sciences casts doubt on the widespread opinion about the benefits of any restrictions in the field of artificial intelligence. The authors of the work argue that fragmented control measures can make the final product much more dangerous than its release without government intervention at all. The analysis is based on a theoretical economic model that describes in detail the behavior of various market participants. The researchers examined the interaction between companies developing fundamental language models and organizations implementing these technologies in specific industries, such as medical diagnostics or customer service.

The key conclusion of the work is that regulation affecting exclusively application companies creates a serious systemic bias. Developers of basic AI systems are starting to save on internal security measures, rightly believing that their partners will take full responsibility. There is a classic stowaway effect, when a large provider shifts the burden of ensuring reliability to highly specialized partners. In conditions of uncertainty, each party instinctively chooses a strategy to minimize its own costs, which ultimately leads to a deterioration in overall safety indicators for the entire ecosystem.

The authors of the study insist that only strict and comprehensive regulation of the entire production chain can radically change this negative dynamic. In their opinion, well-directed rules bring mutual benefits to all participants, significantly improving both the safety of the final product and the long-term economic performance of companies. Regulation is seen not as a profit constraint, but as a reliable tool that eliminates market uncertainty and creates a solid foundation for trust. Currently, this issue is causing active discussions, as different political forces propose opposing approaches to ensuring technological leadership and minimizing potential risks.