2026-W24 · -0.2
This week's index changes only modestly. Research papers highlighted continuing weaknesses in AI safety evaluations, jailbreak resistance and deepfake detection, but most evidence was lab-based rather than real-world harm. A benchmark reporting no successful multi-step prompt-injection attacks against current frontier computer-using agents was an important counterweight. Enterprise availability of frontier models through AWS slightly increased the acceleration signal.
A conventional composite index, not a scientific probability of extinction.