Understanding the First Wave of AI Safety Institutes: Characteristics, Functions, and Challenges

Safety evaluations are the core function of AI Safety Institutes: First-wave AISIs (UK, US, and Japan) focus on testing AI systems across tasks to understand their behavior and identify risks related to cyber, chemical, and biological misuse—providing crucial technical assessments that didn't previously exist in a coordinated way.

These institutes lack regulatory power, which is both a feature and a limitation: While first-wave AISIs are government institutions with clear safety mandates, they cannot enforce rules or mandate compliance, meaning their impact depends on voluntary cooperation from AI developers and other institutions.

International coordination is emerging but fragmented: Seven jurisdictions have now established AI Safety Institutes or similar institutions, but there is "considerable variation between these institutions," suggesting that global AI safety efforts may lack unified standards or approaches despite shared goals.

Three core functions support AI safety work: Research, standards-setting, and international cooperation form the backbone of AISI activities, helping build scientific consensus and foundational knowledge about AI safety—not just conducting evaluations.

The AISI model faces criticism for potential redundancy and narrow focus: Some analysts worry that first-wave AISIs may specialize too narrowly in specific safety areas and duplicate work already being done by existing institutions, raising questions about efficiency and whether this institutional model is the most effective approach.