Crown-International-AI-Safety-Report-2026.md

International AI Safety Report 2026: concise assessment of general-purpose AI. The report reviews recent advances (reasoning systems, inference-time scaling, synthetic data), documents current capabilities across domains (mathematics, software engineering, science, medicine), and highlights a jagged capability profile and an evaluation gap. It analyses three risk categories — malicious use (fraud, influence operations, cyberattacks, biological/chemical misuse), malfunctions (hallucinations, agent brittleness, loss-of-control concerns), and systemic risks (labour market impacts, threats to autonomy) — and outlines mitigation and governance approaches (Frontier AI Safety Frameworks, red-teaming, defence-in-depth, resilience-building). The report emphasises major evidence gaps, uncertain trajectories to 2030, and the need for improved evaluation, transparency, cross-sector coordination, and international cooperation to manage risks while enabling benefits.