Anthropic has enabled Auto Mode as the default setting for Claude Code on paid plans, streamlining workflows but shifting more responsibility onto developers and raising safety questions amid internal and external testing.
Anthropic is making Auto Mode the default for Claude Code on paid plans, a shift that moves the coding assistant further towards acting on behalf of developers rather than waiting for repeated permission prompts. The company says the change is meant to reduce the habit of rubber-stamping routine approvals, while preserving a check for actions the system judges to be risky or irreversible. Anthropic’s own blog says the feature uses a classifier to review tool calls before they run, with users still asked to confirm higher-risk steps.
The timing matters. Anthropic said Auto Mode became generally available on July 10, 2026, after an earlier research preview, while the separate report that highlighted the default change said it would reach Pro, Max and Team users on August 14, with Enterprise customers needing to opt in. A later update from a managed-provider channel also said Claude Code 2.1.207 removed the opt-in requirement for Bedrock, Vertex AI and Foundry, leaving operators to disable Auto Mode explicitly if they want the old behaviour.
Anthropic is backing the rollout with safety claims drawn from internal testing and an outside audit. Trajectory Labs reportedly ran 72 prompt-injection scenarios 10 times each against Claude Code in Auto Mode, producing 720 attempts in total, without a single successful attack against Anthropic’s current models. The same test set against OpenAI’s GPT-5.6 Sol in Codex Auto-Review mode reportedly succeeded 5.83% of the time, according to the report cited by the original article.
The company says 1,053 paid testers used Auto Mode alongside internal red-teaming, and that the feature was at least as safe as manual approvals while improving throughput. Anthropic also pointed to two near-misses: one where the system blocked confidential data from being posted publicly inside the company, and another where it stopped a long-running session from killing about 2,000 processes that would have disrupted GPU training jobs. That makes the trade-off clear. Auto Mode may cut interruptions, but it also shifts more responsibility on to developers to review what the model produces, especially in production environments.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
