Tag
Anthropic introduces a beta server-side fallback feature for its Claude models, Fable 5 and Opus 5, which uses safety classifiers and automatic API retries to enhance reliability.
This paper introduces Regime-Conditional Verification (RCV), a lightweight wrapper that adapts off-the-shelf safety classifiers for large language models by estimating prediction correctness and detecting distribution shift without retraining.
Anthropic provides detailed information on the cyber safety classifiers for Claude Fable 5 and introduces a draft jailbreak severity framework developed with Glasswing, aiming to standardize communication about AI jailbreak risks. The company also launched a HackerOne program for reporting potential cyber jailbreaks.