As AI capabilities surge exponentially in early 2026, safety improvements lag dangerously behind, leaving critical alignment challenges unsolved.

AI capabilities continue exponential growth visible in the METR graph and revenue metrics, with signs of acceleration as AI is used to develop AI faster

Alignment has improved alongside capability gains across multiple measures including spec compliance, but not at a pace matching the raised stakes

Critical safety problems remain unsolved including adversarial robustness, dishonesty, and reward hacking, falling short of reliability standards needed for high-stakes applications

Nicholas Carlini's research highlights the gap between current AI safety standards and what's required for deployment in critical domains