AI Safety Rules Could Backfire, Study Warns

Weak AI safety regulations could paradoxically increase risks, according to a new study published in the Proceedings of the National Academy of Sciences. Researchers from Cornell and Carnegie Mellon University developed a theoretical model to assess effective AI safety regulation.

Their findings indicate that regulations must target primary AI model developers, such as OpenAI and Google, to ensure genuine safety. Focusing solely on companies applying AI technology, like those in medical diagnostics or customer service, may lead developers to reduce their own safety investments, expecting downstream companies to compensate.

This dynamic creates a 'free-riding' behavior where AI developers offload safety burdens. The study proposes that strict, well-placed regulation can benefit all parties by enhancing both product safety and overall utility, fostering cooperation instead of competition.

The researchers liken the situation to a prisoner's dilemma, where cooperation through comprehensive regulation across the AI supply chain is essential for achieving the most favorable outcomes for everyone involved.