UN says AI safeguards can’t wait for certainty

Governments may now mandate AI safety measures that impact your development or deployment processes, even before the exact risks are proven.

The United Nations scientific panel has issued its first major assessment on AI safety, urging governments to implement safeguards for increasingly capable AI agents despite the absence of complete understanding of their risks. This report places AI firmly on the global diplomatic agenda, coinciding with the UN General Assembly and bilateral talks between the US and China on the subject. UN Secretary-General António Guterres has called for international cooperation to mitigate AI threats, emphasizing that a "race to the bottom on AI safety" is unaffordable. The panel, established last year as the UN’s foremost global scientific body on AI, advocates for increased resources and international coordination on AI safety and accountability, even as individual nations pursue distinct legal frameworks. A key recommendation is to apply the precautionary principle, which dictates that serious or irreversible harm should be addressed without waiting for absolute scientific certainty regarding its likelihood or causation. This principle, previously applied in environmental and public health policy, is deemed appropriate for managing AI’s potential loss-of-control risks.

The UN panel's emphasis on the precautionary principle signals a potential shift in how AI risks will be regulated globally, moving from reactive measures to proactive interventions. This approach prioritizes mitigating potential catastrophic or irreversible harm, even in the face of scientific uncertainty, which could lead to more stringent and anticipatory safety requirements for AI development and deployment. Companies may face increased pressure to implement robust safety protocols and demonstrate risk mitigation strategies before deploying advanced AI systems.

This development could accelerate the divergence in AI regulatory landscapes, with nations adopting varying legal approaches but united under a precautionary umbrella. It also highlights the growing interconnectedness of AI and international diplomacy, as seen with concurrent US-China discussions. For developers and businesses, this means navigating a complex and evolving global regulatory environment where the burden of proof for safety may increasingly fall on them, necessitating a proactive and adaptable approach to AI governance.