2026-W23 · +0.3
This week’s dossier points to a small increase in AI risk signals, mainly from research showing weaknesses in monitoring, jailbreak resistance, and distributed agent misuse defenses. The strongest capability concern is a demonstrated distributed cyberattack scaffold that can evade standard single-context monitoring. Real-world incidents remain in the recurring fraud range, with deepfake scams causing individual harm but no systemic effects. Governance evidence is thin, so that score is kept nearly unchanged.
A conventional composite index, not a scientific probability of extinction.