Both the 2023 and 2026 AI safety commitments share the challenge of stating goals only in broad terms without explaining how success should be measured.