Starting August 14, 2026, Anthropic will set "auto mode" as the default configuration for new Claude Code sessions across Pro, Max, and Team plans. This strategic shift follows internal testing and third-party evaluations aimed at addressing risks like prompt injection and data exfiltration. In a study involving 1,053 paid testers, researchers found that humans only refused a single dangerous command in 13.6% of instances due to confirmation fatigue; in contrast, Claude Code's auto mode successfully blocked 89% of those harmful actions. Furthermore, an independent evaluation conducted by Trajectory Labs tested 72 indirect prompt injection scenarios against Claude Fable 5, Opus 5, and Sonnet 5. As of July 17, 2026, none of the 720 attack attempts succeeded against these models while running in auto mode. Despite these results, security concerns remain regarding complex attack vectors, such as malicious third-party packages that use seemingly legitimate commands like "uvx fetch-model-files" to exfiltrate data, posing a challenge to current agentic security frameworks.