AI cooperation stalls as countries disagree on which risks matter

In short: Countries are struggling to work together on AI rules because they do not agree on which AI risks are most important and how severe they are.

What's going on

Many governments say AI risks can cross borders, but turning that into shared rules has been slow. Recent work from groups like the UK government, the UN, Brookings, and Chatham House points to the same roadblock. Countries first need shared definitions of AI risks, plus shared ways to measure them, before deeper cooperation becomes realistic.

Brookings has argued that cooperation would be easier if countries could clearly identify and classify risks, weigh benefits against harms, and agree on when a risk is too high to manage. This is similar to how countries handle public health threats. It is hard to coordinate a response if everyone uses different definitions for what counts as an emergency.

Progress so far has been limited and focused. At the 2024 Seoul AI summit, countries agreed to work on shared thresholds for what counts as “severe AI risks,” including scenarios where very capable AI systems could help create biological or chemical weapons or operate in ways that evade human oversight.

The UN has also said global coordination is necessary, but still described AI governance as an ongoing challenge. Chatham House and other researchers add that competition between countries, weak international institutions, and gaps between governments and private companies make binding agreements difficult.

What to watch

Expect more narrow agreements, like shared testing of “frontier” AI models (the most capable systems) and information sharing, rather than a single global AI rulebook. A bigger shift would likely require countries to agree on a common priority list of risks, whether that is security threats, privacy, bias, or market power.

Source: NYTimes