Anthropic makes Claude Code's Auto Mode the default, citing human approval fatigue
Starting August 14, Pro, Max, and Team users get AI-driven permission review by default after tests showed humans rubber-stamp 97% of prompts.
Conversation activity · last 3 days peak 11/hr
Summary, timeline and people extracted by Claude from 41 items across 7 sources · 17h ago. Quotes are verbatim.
Anthropic announced that Auto Mode — a classifier that reviews Claude Code's shell commands and blocks dangerous ones instead of asking for human approval — will become the default permission setting for Pro, Max, and Team plans starting August 14, 2026. The company's own testing found that human reviewers caught only 13.6% of deliberately inserted dangerous commands (dropping to 5% after 50 prompts) versus 89% for the auto mode classifier, and cited a third-party evaluation claiming zero successful prompt-injection attacks out of 720 attempts. Commentators, including security researcher Simon Willison, welcomed the safety framing but questioned whether the claims fully address indirect prompt injection risks from malicious third-party content.
- Auto mode becomes the default for Claude Code Pro, Max, and Team plans on August 14, 2026; it stays opt-in for Enterprise and API/cloud users for now, with default rollout there planned within a month.
- Anthropic says users currently approve 97% of permission prompts reflexively; in a 1,053-tester study, humans caught only 13.6% of a planted dangerous command (5% after 50 prompts) versus 89% caught by the auto mode classifier.
- New hardening includes hard denies on data exfiltration, org-level custom deny rules, git-status checks before destructive commands, and a third-party (Trajectory Labs) test reporting 0 of 720 indirect prompt-injection attempts succeeded against Claude models in auto mode.
- Independent commentators like Simon Willison find the results promising but want more independent confirmation, noting scenarios like malicious third-party packages that current auto mode may not stop.
How it unfolded
-
Event Default rollout takes effect
Auto mode is scheduled to become the default permission setting for Claude Code Pro, Max, and Team users, with Enterprise and API defaults expected within about a month.
-
Report TechCrunch covers the rollout and Cherny's endorsement
TechCrunch reports the default change and quotes Claude Code Head Boris Cherny saying his team has used auto mode exclusively for months.
“The team and I use Auto mode exclusively, and have been for many months. I couldn’t imagine going back to permission prompts!”
Boris Cherny · Bluesky ↗ -
Analysis Simon Willison scrutinizes prompt-injection claims
Willison highlights Anthropic's claim that a third-party evaluation (Trajectory Labs) found zero successful attacks out of 720 prompt-injection attempts, but says he wants independent confirmation and raises unresolved attack scenarios.
“I would love to believe that Anthropic have indeed solved this problem for Claude Code users.”
Simon Willison · Hacker News ↗ -
Analysis The New Stack details the '97% problem'
The New Stack reports Anthropic's findings that humans approve 97% of permission prompts reflexively and describes new hardening measures like hard denies on data exfiltration and git-status checks before destructive commands.
“While most prompts are likely for safe, routine commands, an approval rate that high suggests many users are clicking through reflexively rather than reviewing each command”
Anthropic · Hacker News ↗ -
Event Anthropic announces default switch for Aug 14
Claude Code's official account tweets that auto mode will become the default permission mode for Pro, Max, and Team users starting August 14.
“Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users. Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual…”
@claudedevs · X ↗ - 23 weeks quiet
-
Event Auto mode first introduced as opt-in test
Anthropic unveiled a test version of auto mode in March 2026, pitched as a way to balance agent speed with human control.
What people are saying verbatim
“While most prompts are likely for safe, routine commands, an approval rate that high suggests many users are clicking through reflexively rather than reviewing each command”
Anthropic, Company statement · The New Stack ↗ · Aug 7
“The team and I use Auto mode exclusively, and have been for many months. I couldn’t imagine going back to permission prompts!”
Boris Cherny, Claude Code Head · TechCrunch / X ↗ · Aug 8
“We’re going to publish some evals in the coming weeks, but we’ve pretty much mitigated every attack.”
Cat Wu, Anthropic, Claude Code team · Simon Willison's Weblog ↗ · Aug 7
“we should have called this post "defeating the lethal trifecta"”
Thariq Shihipar, Anthropic, Claude Code team · Twitter, quoted by Simon Willison ↗ · Aug 7
“I would love to believe that Anthropic have indeed solved this problem for Claude Code users.”
Simon Willison, Independent commentator · Simon Willison's Weblog ↗ · Aug 7
“Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users. Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual approval caught 14%.”
@claudedevs, Official Claude Code account · X (Twitter) ↗ · Aug 6
“Anthropic will make Auto Mode the default in Claude Code for Pro, Max, and Team plans starting August 14, citing a classifier that catches 89% of dangerous commands versus 13.6% for humans. Developers may spend less time writing code and more time monitoring AI output.”
@sipirtu.com, Bluesky user · Bluesky ↗ · Aug 8
Voices from the web unedited
-
Anthropicが「Claude Code」の自動モードをデフォルト設定に変更。AIが自律的にコード実行やコマンド操作を行うようになり、開発効率が飛躍的に向上します。コーディングの自動化がまた一段と加速しそうですね。 #AI #Claude
-
For small projects where I don't wish to interact with Claude so much, I've started using Anthropic's sandbox runtime utility "srt":https://github.com/anthropic-experimental/sandbox-runtimeThis is combined with "auto" mode.Seems to work so far. I've manually checked various things, rw access, access to sensitive folders/files etc.So far, I've only…
-
We're making this change for two reasons: 1. In our testing, auto mode matched or beat manual permission review on every safety measure we tracked. 2. It makes long-horizon work more viable. Claude runs longer between interruptions, so you can run multi-hour tasks in the
-
Programming with Claude Code will soon require even less human oversight. https:// techcrunch.com/2026/08/09/anth ropic-is-turning-claude-codes-auto-mode-on-by-default/?utm_source=dlvr.it&utm_medium=mastodon
-
Anthropic is making Claude Code auto mode the default, a clear sign that agentic coding is moving from opt in to mainstream. - techcrunch.com/2026/08/09/a... #AI #Anthropic #ClaudeCode
-
Besides the debate about whether this is "safer" than manual human review, I have a slightly different problem.Very often, when I'm running Claude in manual review mode, it will attempt to do things which are not "dangerous" but are misaligned with what I want it to do. Maybe I'm fighting the model here but for example, when orchestrating other…
-
Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users. Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual approval caught 14%.
-
Anthropic will make Auto Mode the default in Claude Code for Pro, Max, and Team plans starting August 14, citing a classifier that catches 89% of dangerous commands versus 13.6% for humans. Developers may spend less time writing code and more time monitoring AI output. — Source: The Decoder AI
-
I have been running Claude Code using its built-in /sandbox feature with this config: "sandbox": { "enabled": true, "failIfUnavailable": true, "autoAllowBashIfSandboxed": true, "allowUnsandboxedCommands": false, "filesystem": { "allowWrite": [ "." ], "denyRead": [ "~/*" ], "allowRead": [ ".", // a few more dirs ] } }, "defaultMode": "auto" But…
-
Programming with Claude Code will soon require even less human oversight. #Techcrunch #breakingnews #news #othur