Anthropic introduces a new Auto Mode for Claude Code that uses model-based classifiers to automatically block dangerous actions while executing safe operations without approval prompts, combining an input-side prompt injection probe with an output-side transcript classifier.