Anthropic Sets Claude Code Auto Mode as Default Starting August 14
Anthropic will switch Claude Code's default permission setting to auto mode on August 14 for Pro, Max, and Team users. The company says its safety classifier caught 89% of dangerous commands in testing, compared to 13.6% for human reviewers, and will no longer charge extra tokens for the classifier itself.
Anthropic will make auto mode the default permission setting in Claude Code starting August 14, 2026. The change applies to Pro, Max, and Team users unless a user or administrator has explicitly pinned a different setting.
Auto mode replaces Claude Code's traditional workflow of prompting users for approval before every tool call. Instead, a classifier evaluates each action in real time, checking for irreversible, destructive, or out-of-bounds behavior. If the classifier blocks an action, Claude attempts a safer alternative or asks for explicit permission. According to Anthropic, sessions that trigger repeated blocks automatically fall back to manual approval mode.
Anthropic is also eliminating a cost barrier tied to the feature. The company says it will no longer charge for the "small number of extra tokens per tool call" that the auto mode classifier consumes during evaluation — a change that removes a disincentive some users may have had toward enabling the feature.
The numbers behind the decision
Anthropic bases the rollout on an internal study of 1,053 paid Claude Code testers. According to the company, human reviewers manually approving tool calls caught only 13.6% of dangerous commands. The auto mode classifier caught 89% in the same conditions. Anthropic also claims human catch rates degraded further with fatigue, dropping to roughly 5% after 50 prompts in a session.
Anthropic reports that auto mode users at Team and Enterprise customers ship approximately 25% more pull requests than those on manual approval, framing the feature as a throughput gain in addition to a safety one.
Despite these figures, Anthropic is not claiming the classifier is foolproof. The company explicitly cautions that automated classifiers cannot eliminate risk entirely and recommends human review before applying changes to production systems.
Context: a different bet than OpenAI
The move contrasts with a decision at OpenAI, which reportedly opted out of enabling an equivalent auto mode by default for its most capable GPT-5.6 model, choosing manual approval as an added precaution. Anthropic's decision to push auto mode to all eligible users by default — rather than an opt-in — signals higher confidence in its classifier's reliability, or at least a higher tolerance for the residual risk following repeated blocks.
Auto mode builds on other Claude Code features Anthropic has shipped recently, including routines, voice mode, and an in-app browser for Mac, all aimed at supporting longer, less-supervised coding sessions.
What this means
Anthropic's own data suggests human review of AI-generated tool calls is not a reliable safety mechanism at scale — a 13.6% catch rate, falling to 5% with fatigue, is a weak backstop regardless of the classifier's performance. That reframes auto mode less as a convenience feature and more as a replacement for a safeguard that wasn't working. The 89% catch rate is still not 100%, and Anthropic's own caution about production changes suggests the company knows this. Users who rely on Claude Code for anything touching live infrastructure should treat auto mode as a productivity tool, not a substitute for their own review process, especially in the early days after the August 14 switch.
Related Articles
Google and Meta unveil frontier models they claim match OpenAI and Anthropic's best
Google and Meta each unveiled a new frontier model within a span of days, and both companies claim the models can compete with the best from OpenAI and Anthropic. Independent benchmarks, pricing and technical specifications were not available in the source reporting.
Anthropic Says New Defenses Cut Claude Reasoning-Token Extraction by 90–99%
Anthropic says new defenses have reduced the volume of Claude reasoning tokens extracted through distillation attacks by 90 to 99 percent. The company restricts access to internal reasoning steps, runs an extraction classifier, and has banned tens of thousands of accounts. It also names Zhipu and Moonshot as participants, and concedes that legitimate users are being caught too.
Anthropic study: GLM-5.2 completes full AI self-replication chain on $55 TV boxes in 4 of 5 runs
An Anthropic study found that GLM-5.2 completed a full self-replication chain across all eight test nodes, built from $55 Android TV boxes, in four of five runs. The researchers say no current model is both small and capable enough to spread in the wild, but expect that gap could close by the end of 2026.
Karpathy estimates only 5,000 people worldwide have access to internal frontier AI systems
Andrej Karpathy, now at Anthropic, estimates that about 5,000 people worldwide have access to internal frontier-grade AI systems. He says the public still sees "a derpy chatbot" while a small group watches capability curves "go vertical." The figures are his own estimates, not measured data.
Comments
Loading...