Claude Blog 采集 (2026-08-11)¶
共采集 3 篇文章
📋 文章索引¶
- Auto mode is now the default in Claude Code for Pro, Max, and Team plans - Aug 07, 2026 (评分: 9.5)
- Running auto mode in production - Aug 07, 2026 (评分: 9.5)
- Run Claude Code sessions on your own compute - Aug 06, 2026 (评分: 9.0)
Auto mode is now the default in Claude Code for Pro, Max, and Team plans¶
来源: Claude Blog 发布日期: Aug 07, 2026 采集时间: 2026-08-11 价值评分: 9.5/10 正文字数: ~7055 字符
摘要¶
Claude Code will soon run auto mode by default for Pro, Max, and Team plans, enabling longer-running autonomous work, and catching more dangerous commands.
正文内容¶
Claude Code will soon run auto mode by default for Pro, Max, and Team plans, enabling longer-running autonomous work, and catching more dangerous commands than manual review in our testing.
We're making auto mode the default in Claude Code. Starting on August 14, new sessions on Pro, Max, and Team plans will run in auto mode. If you've already set a different default yourself, you may get a one-time prompt asking whether you want to switch to auto mode. If you have a pinned default, nothing changes for you. The auto mode classifier uses a small number of extra tokens per tool call, and we're no longer charging Claude Code users on Pro, Max, and Team plans for that classifier overhead, effective today. Auto mode remains opt-in for now on Claude Enterprise, the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform, and Microsoft Foundry, giving admins time to review the change. In the coming month, working with our cloud partners, we plan to make it the default across all of these and no longer charge for classifier overhead. In the meantime, Enterprise admins can make Claude Code's auto mode the default through managed settings. Auto mode is designed to balance users’ desire not to be interrupted with a system that helps avoid harmful actions: instead of prompts, it routes each tool call through a classifier targeted at blocking actions that are irreversible, destructive, or aimed outside your environment. When the classifier blocks something, Claude usually finds a safer way to proceed on its own or asks you directly for the go-ahead; if it can't make progress—three blocks in a row, or twenty across a session—Claude Code falls back to manual approvals. We spent the last several months testing whether auto mode is as safe or safer than an average user clicking through prompts. We ran internal red-teaming, third-party red-teaming and prompt-injection evaluations, a controlled study with 1,053 paid testers, and analysis of real production sessions. On every measure we tested, auto mode matched or outperformed manual review. Auto mode also lets Claude work autonomously for longer stretches. This makes models built for long-running work, like Claude Opus 5, more practical to leave running for hours on large tasks. Reducing overhead for users also increases output. Among Teams & Enterprise adopters, auto mode users ship about 25% more PRs. Unblocking Claude allows tasks to run longer uninterrupted and get more work done. Teams at Adobe, Nuro, Gusto, and Garner Health already run auto mode as their production default. Below, we share the safety data and customer results motivating the change, and how to set a different default if you prefer. Comparing manual review to auto mode Data suggests that manual review can become habitual: users approve 97% of permission prompts in Claude Code. 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. These prompts ask developers to make dozens or hundreds of important security decisions every day, often in the middle of projects, which places the review burden on users and increases the chance that something important slips through the cracks. Data also suggests that users more frequently scrutinize and push back on other types of dialogues: for example, when Claude presents a plan for approval, users reject 39% of them. But for individual permissions requests, the rejection rate is only 3%. The same pattern shows up in settings files. As of June 2026, 49.5% of active CLI users have manually created a Bash allow-rule—5% allow any shell command outright, and another 43% have interpreter rules like Bash(python:) or Bash(node:) that are essentially equivalent in practice—and that share is growing roughly 5 percentage points every 5 weeks. Beyond allow-rules, 62% of users have used bypassPermissions or clicked “don't ask again” on Bash, and 25% of interactive sessions start in bypass permissions mode. Permission rules still fire before the classifier in auto mode, except for allow rules broad enough to grant arbitrary code execution (e.g. python:*). These arbitrary rules are set aside while in auto mode, since they would let commands skip the classifier entirely. Settings files aren't modified, and the rules apply again the moment you switch to another mode. Improving safety with auto mode Research conducted internally and through several third parties suggests that auto mode is safer than how a typical user engages with permission prompts today. We tested auto mode against what developers actually approve when they're clicking through permission prompts. Across each of these measures—internal red-teaming, third-party penetration testing, and analysis of real-world sessions—auto mode matched or outperformed human approval. Auto mode outperforms manual permissions on safety
Learn how these customers are running auto mode in production . Getting started For Pro, Max, and Team users: if you haven’t set a default permission mode, you’ll receive an in-product notice and new sessions will start in auto mode automatically. If you've set a different default, you may see a one-time prompt asking if you’d like to switch your default to auto mode. If your Team admin has set a default in managed settings, nothing changes for you. For Enterprise users and users who access Claude Code via the Claude API, auto mode remains opt-in for now. We plan to make auto mode the default in the coming month, and we’ll notify Enterprise admins before we do. To switch modes, press Shift+Tab in the CLI or use the mode dropdown on the desktop app. Admins can pin an org-wide default with defaultMode in managed settings , or turn auto mode off entirely with disableAutoMode. Finally, while we believe auto mode reduces risk for most users, it relies on classification systems and therefore does not eliminate risk. For high-stakes changes to production infrastructure, we still recommend reviewing Claude's actions yourself. See the auto mode docs for full configuration instructions. This article was written by Conner Phillippi, with contributions by Nicholas Carlini, Isaac Fung, John Hughes, Alex Isken, Shawn Moore, Javier Rando, and Molly Vorwerck. The authors would also like to thank Yacine Azmi, Chandler Bair, Kefan Chen, Boris Cherny, Ian Grunert, Lydia Hallie, Alex Kleiman, Lauren Polansky, Deon Poncini, Robert Schonberger, Marie Vachovsky , Qing Wang, Cat Wu, Daniel Xu, and Alice Zhao. 1 We evaluated Claude Code v2.1.205 and Codex v0.144.5. OpenAI released a new version of Auto-review last week that could change the results.
Explore more product news and best practices for teams building with Claude.
Transform how your organization operates with Claude
Product updates, how-tos, community spotlights, and more. Delivered monthly to your inbox.
Please provide your email address if you'd like to receive our monthly developer newsletter. You can unsubscribe at any time.
采集自 Claude Blog，由 collect_claude_blog.py 自动采集
Running auto mode in production¶
来源: Claude Blog 发布日期: Aug 07, 2026 采集时间: 2026-08-11 价值评分: 9.5/10 正文字数: ~9400 字符
摘要¶
How customers like Nuro, Gusto, and Garner Health use auto mode to drive safer, longer running coding workflows.
正文内容¶
How the teams at Nuro, Gusto, and Garner Health use auto mode to balance speed and safety at production scale.
Auto mode is now the default setting in Claude Code. Instead of asking you to approve every command an agent wants to run, a classifier evaluates each action and blocks ones that look potentially harmful. Auto mode’s design resolves a common agentic coding tradeoff: speed vs. safety. Reviewing every command keeps a human in the loop, but once sessions stretch to hours or multiply in parallel, that oversight becomes the bottleneck. Skipping permission checks entirely is faster—and it’s also how prompt injection, scope drift, and the occasional deleted production resource get through. Auto mode closes most of that gap. In internal evaluations, the classifier caught more dangerous actions than developers did when clicking through permission prompts by hand, and its performance held up under third-party red-teaming. And because sessions pause less often, Claude works 9x longer between interruptions than under the previous default—across all Claude Code usage. To see how auto mode holds up in production, we spoke with teams at Nuro, Gusto, and Garner Health about how and why they use auto mode as their daily driver to balance speed with safety in their production environments. Powering longer running autonomous agents at Nuro Nuro, the physical AI company developing universal Level 4 autonomous driving technology, adopted Claude Code in late 2025, and by March it was the most popular agentic coding tool at the company. Before auto mode shipped, staff software engineer Kai Zhou had already started prototyping an internal stand-in: a hook that sent each pending action to a small model, auto-approved the routine 90 percent of the time, and routed anything sensitive to Slack for a human to review. The prototype answered a real tension: engineers hated babysitting approval prompts, but from a company security and legal standpoint, skipping permissions outright was too dangerous to sanction. When auto mode shipped, Kai shelved the side project. Today, Kai runs auto mode for everything he writes. "I don't want to sit there and click approve all the time," said Kai. "I use auto mode for 100 percent of my coding work. Most of the time, I open three or four sessions running auto mode in parallel and just check in when I need to.” The exception is work that touches other teams. For instance, when Claude Code reviews a Pull Request on his behalf, Kai switches back to interactive mode and reviews each one before it goes out. Auto mode doesn’t run unconstrained, either. Nuro leans heavily on skills , and engineers deny the most dangerous commands, like recursive deletes, outright in their settings. The classifier makes its judgment calls inside those guardrails. The bigger auto mode unlock, however, has been the ability to kick off work that keeps running after engineers are done for the day. Specifically, Kai’s team uses auto mode to power long-running research agents that hill-climb the evaluation metrics behind its autonomous-driving stack: tasks with a clear, measurable signal an agent can iterate against on its own. Overnight, an agent can study false negatives flagged by the evaluation suite, draft a proposal, run experiments, and keep iterating on the results. The approach extends to any task with a clear evaluation method—another team at Nuro uses it to shrink the memory footprint of a specific binary—because the metric itself tells the agent whether it’s improving or regressing. "The other day, I kicked off an agent at 10 p.m. and it kept running until 5 a.m.—and it gave me three PRs in the morning," Kai said. "I think it's pretty impressive. Only auto mode enables this kind of workload." Shipping PRs faster and safer at Gusto At Gusto, a leading SMB technology company, the move to auto mode started as a proactive security upgrade. Martin Emde, who works on the company's AI Dev Tools team, had watched permission fatigue slow the team down. Auto mode gave them the same velocity without sacrificing control or security, and since adoption took hold across engineering, the overall permissions burden has noticeably declined. Martin has kicked off 2,425 Claude Code sessions since December, with auto mode as his daily driver. Cross-repo work that used to stall on folder-access approvals now runs uninterrupted, and unattended jobs, like compiling daily notes from GitHub, Slack, and Jira, run on their own. In his team’s own analysis, roughly 10% of session transcripts since mid-May 2026 included an auto mode denial, evidence the classifier is doing real work without dragging on legitimate tasks. “Auto mode gave us a safer balance between speed and control," Martin said. "We were able to remove the repeated prompts and increase productivity without compromising safety. We can see that auto mode blocks at the right time, which gives us the confidence to move quickly." Chad Kunsman, a member of Gusto’s AIT Cloud Engineering team, came to the same conclusion from the other direction. His work—endpoint investigations, log audits, connector management, doc ingestion across a stack of MCP servers—runs in short, twenty-minute bursts rather than overnight marathons. He wasn't looking for longer runs; he wanted the hands-off pace of bypass permissions without the exposure of a bad prompt, or a prompt injection, slipping through. "Given the protection against prompt injection, and the way it checks that what you're doing actually lines up with what you asked for, it's the better choice than bypass permissions and far faster than permission prompts," said Chad. On the rare occasions the classifier does step in, Chad says it's on the mark. "When it stopped me, it made sense and explained why. It was drifting from what I'd originally asked, and it checked in. It wasn't off base at all." Chad still steps out of auto mode for his most sensitive work. When a session has its teeth into production infrastructure—Terraform, AWS, direct POST calls against live APIs—he switches to accept edits and verifies each tool call by hand. “You have to weigh the amount of time you’re saving against what it could reasonably make a mistake on, and how catastrophic that would be,” he said. “Ultimately, you’re still responsible for what happens.” That judgment operates inside a broader defense-in-depth setup: Gusto routes its MCP traffic through a governed proxy layer with tool guards and prompt inspection, so agents work with tightly scoped permissions before auto mode ever weighs in. Accelerating the software development lifecycle (SDLC) at Garner Health Garner Health, the healthcare technology company, rolled out Claude Code in February to all 550 employees across every function. The tool is wired into all the core systems including Salesforce, Zendesk, and Snowflake, and employees are encouraged to spend about two hours a week automating the most repeatable parts of their job. Before auto mode, that scale came with overhead. Evan Magnussen, Garner's platform engineering manager, describes permission management as a tedious cycle of hand-curating approved command lists and watching piped commands get rejected. Today, Evan and most of his colleagues use auto mode in every session, from researching the codebase to managing external integrations through MCP. “We've built out a standardized software development lifecycle for the entire engineering organization that is really only possible because of auto mode,” Evan said. “Employees view it as a weight off their shoulders. They don’t have to monitor their agents for hours on end anymore." That lifecycle runs as a plugin of standardized skills. An agent picks up a task, explores the context it has access to, commits context files to the repository, runs what Evan calls “antagonistic research” to pressure-test its own assumptions, and then moves on to implementation—pausing for a human only when it needs context it can’t find on its own. The research-heavy stages, Evan notes, weren’t possible before auto mode. Out of the box, the classifier has needed little tuning. Evan’s one adjustment mirrors Kai’s at Nuro: he configured auto mode not to approve actions that communicate with other people, like sending Slack messages or emails. “I personally don’t like Claude to just act on my behalf when I’m communicating with another person,” he said. Teams working on core intellectual property—the most skeptical of skipping permissions before auto mode—learned to tune the classifier’s injected prompts to be more or less permissive for their work. His advice for other enterprises rolling it out? Lean in and build the right controls so that you can empower engineers while ensuring safe deployment. “If we were to say, everyone go build your own workflows, and we have no telemetry, that would be very dangerous,” Evan said. “Because we have the telemetry, because we’ve built out workflows that are relatively standard, we have much more confidence.” Get started with auto mode in Claude Code.
Explore more product news and best practices for teams building with Claude.
Transform how your organization operates with Claude
Product updates, how-tos, community spotlights, and more. Delivered monthly to your inbox.
Please provide your email address if you'd like to receive our monthly developer newsletter. You can unsubscribe at any time.
采集自 Claude Blog，由 collect_claude_blog.py 自动采集
Run Claude Code sessions on your own compute¶
来源: Claude Blog 发布日期: Aug 06, 2026 采集时间: 2026-08-11 价值评分: 9.0/10 正文字数: ~3566 字符
摘要¶
Self-hosted environments are in public beta. Run Claude Code sessions on your own infrastructure, inside your network and next to your internal services.
正文内容¶
Now in public beta, self-hosted environments let you run Claude Code sessions on your own infrastructure. Start a session from the web, mobile, desktop, or a routine, and it runs inside your network, next to your internal services, toolchains, and security controls, rather than on Anthropic-hosted infrastructure. For most enterprises, we strongly recommend our hosted offering for operational simplicity with no infrastructure to run or maintain. Self-hosted environments are for teams whose network, tooling, or compliance requirements call for keeping agent execution on infrastructure they control. If you go this route, plan to staff engineering to own setup and ongoing maintenance. Why self-host We saw organizations in our preview program adopt self-hosted environments for a few key reasons: Network access: sessions run inside your network and can reach internal services, databases, and registries without exposing them to the public internet Customizability: pre-install compilers, SDKs, and internal CLIs in your environment so every session starts ready to build Compliance: source code and build artifacts stay on infrastructure you control
Data stays on your infrastructure Repository checkouts, build artifacts, secrets, and any files a session creates or modifies all stay on infrastructure you provision. The conversation itself, including prompts, responses, and tool results (which can include code that Claude reads), is sent to Anthropic for inference, and the session transcript is stored so a session can be picked up from any surface. How it works When using self-hosted environments, you deploy a set of runners . These long-lived processes pick up sessions and start a Claude Code process for each session . Runners come in two modes. Fixed: you keep a set number running and sessions are distributed across them. On-demand: an orchestrator watches for queued sessions, starts a runner as sessions arrive, and stops them when work finishes so capacity tracks demand. Runners can serve more than one session, but each session runs in its own checkout, so work stays isolated between developers and accounts. Sessions from every supported surface route to the same environment, so you set it up once and it works wherever your team starts a session. Note : Self-hosted environments differ from Remote Control , which lets developers continue sessions running on their own machines from a phone or browser. Sessions using Remote Control end when that machine stops running the session and are tied to the user who ran claude , whereas self-hosted environments run sessions on shared infrastructure your platform team operates and can be used by any user. Getting started Self-hosted environments are available in public beta to organizations on Claude Team and Enterprise plans. They are off by default and not available for organizations using ZDR. Plan on a platform, developer experience, or developer productivity team owning setup and ongoing operation, including building and maintaining the runner image, updating runners, and running the orchestrator if you use on-demand mode. See the documentation to learn more. Share feedback via GitHub or through your Anthropic account team.
Explore more product news and best practices for teams building with Claude.
Transform how your organization operates with Claude
Product updates, how-tos, community spotlights, and more. Delivered monthly to your inbox.
Please provide your email address if you'd like to receive our monthly developer newsletter. You can unsubscribe at any time.
采集自 Claude Blog，由 collect_claude_blog.py 自动采集