GoKawiilAnthropic has begun previewing "auto mode" inside of Claude Code. The company describes the new feature as a middle path between the app's default behavior, which sees Claude request approval for every file write and bash command, and the "dangerously-skip-premissions" command some coders use to make the chatbot function more autonomously.
With auto mode enabled, a classifier system guides Claude, giving it permission to carry out actions it deems safe, while redirecting the chatbot to take a different approach when it determines Claude might do something risky. In designing the system, Anthropic's goal was to reduce the likelihood of Claude carrying out mass file deletions, extracting sensitive data or executing malicious code.
Of course, no system is perfect, and Anthropic warns as such. "The classifier may still allow some risky actions: for example, if user intent is ambiguous, or if Claude doesn't have enough context about your environment to know an action might create additional risk," the company writes.
Advertisement Advertisement
Advertisement
Anthropic doesn't mention a specific incident as inspiration for auto mode, but the recent 13-hour AWS outage Amazon suffered after one of the company's AI tools reportedly deleted a hosting environment, was probably front of mind for the company. Amazon blamed that specific incident on human error, saying the staffer involved in the incident had "broader permissions than expected."
Team plan users can preview auto mode starting today, with the feature set to roll out to Enterprise and API users in the coming days.