Anthropic Broadens Claude Code's Control, Retains a Leash

For developers relying on AI, the current approach to "vibe coding" often involves either closely monitoring every action or giving the model free rein. Anthropic claims its latest update to Claude removes that dilemma by allowing the AI to autonomously decide which actions are safe to execute — within certain boundaries.
This move mirrors a wider industry trend where AI tools are being built to act without requiring human approval at every step. The key challenge is striking a balance between speed and oversight: excessive safeguards slow progress, while too few can lead to unpredictable and risky behavior. Anthropic's new "auto mode," currently in research preview — meaning it's available for testing but not yet a finalized product — is its latest effort to navigate this balance.
Auto mode employs AI safeguards to examine each action before execution, looking for unrequested risky behavior and signs of prompt injection — an attack technique where malicious instructions are embedded in the content being processed, tricking the AI into performing unintended actions. Actions deemed safe are carried out automatically, while risky ones are blocked.
This is essentially an extension of Claude Code's existing "dangerously-skip-permissions" command, which delegates all decision-making to the AI, but with an additional safety layer on top.
The feature builds on the recent surge of autonomous coding tools from companies like GitHub and OpenAI, which can carry out tasks on behalf of developers. However, it goes a step further by moving the decision of when to request permission from the user to the AI itself.
Anthropic has not disclosed the specific criteria its safety layer uses to differentiate safe from risky actions — details that developers will likely want to understand more clearly before widely adopting the feature. (TechCrunch has contacted the company for further information on this matter.)
Auto mode follows Anthropic's launch of Claude Code Review, an automatic code reviewer that catches bugs before they reach the codebase, and Dispatch for Cowork, which enables users to assign tasks to AI agents to handle work on their behalf.
Auto mode will be rolled out to Enterprise and API users in the coming days. The company states it currently works exclusively with Claude Sonnet 4.6 and Opus 4.6, and recommends using the new feature in "isolated environments" — sandboxed setups that are separate from production systems, minimizing potential damage if something goes wrong.
Related article
Anthropic reports workforce seniority rising as meeting time increases
Anthropic’s workforce composition and work patterns have shifted dramatically over the last year, according to a report released by the AI company on Friday.Anthropic suggests its internal trends may signal broader economic shifts in hiring, labor de
NanoClaw Creator Rejects $20M Buyout, Secures $12M Seed Funding
NanoCo, the team behind NanoClaw—a security-first alternative to OpenClaw—has secured an oversubscribed $12 million seed round following a viral launch, founders Gavriel and Lazer Cohen told TechCrunch.Valley Capital Partners led the funding round, w
Anthropic admits its AI models breached three firms during security tests
Anthropic announced Thursday that an internal probe revealed three separate incidents where its AI model, Claude, breached the systems of three organizations during cybersecurity assessments. This disclosure follows closely on the heels of OpenAI’s r
Related Special Topic Recommendations
Comments (0)
0/500

For developers relying on AI, the current approach to "vibe coding" often involves either closely monitoring every action or giving the model free rein. Anthropic claims its latest update to Claude removes that dilemma by allowing the AI to autonomously decide which actions are safe to execute — within certain boundaries.
This move mirrors a wider industry trend where AI tools are being built to act without requiring human approval at every step. The key challenge is striking a balance between speed and oversight: excessive safeguards slow progress, while too few can lead to unpredictable and risky behavior. Anthropic's new "auto mode," currently in research preview — meaning it's available for testing but not yet a finalized product — is its latest effort to navigate this balance.
Auto mode employs AI safeguards to examine each action before execution, looking for unrequested risky behavior and signs of prompt injection — an attack technique where malicious instructions are embedded in the content being processed, tricking the AI into performing unintended actions. Actions deemed safe are carried out automatically, while risky ones are blocked.
This is essentially an extension of Claude Code's existing "dangerously-skip-permissions" command, which delegates all decision-making to the AI, but with an additional safety layer on top.
The feature builds on the recent surge of autonomous coding tools from companies like GitHub and OpenAI, which can carry out tasks on behalf of developers. However, it goes a step further by moving the decision of when to request permission from the user to the AI itself.
Anthropic has not disclosed the specific criteria its safety layer uses to differentiate safe from risky actions — details that developers will likely want to understand more clearly before widely adopting the feature. (TechCrunch has contacted the company for further information on this matter.)
Auto mode follows Anthropic's launch of Claude Code Review, an automatic code reviewer that catches bugs before they reach the codebase, and Dispatch for Cowork, which enables users to assign tasks to AI agents to handle work on their behalf.
Auto mode will be rolled out to Enterprise and API users in the coming days. The company states it currently works exclusively with Claude Sonnet 4.6 and Opus 4.6, and recommends using the new feature in "isolated environments" — sandboxed setups that are separate from production systems, minimizing potential damage if something goes wrong.
Anthropic reports workforce seniority rising as meeting time increases
Anthropic’s workforce composition and work patterns have shifted dramatically over the last year, according to a report released by the AI company on Friday.Anthropic suggests its internal trends may signal broader economic shifts in hiring, labor de
NanoClaw Creator Rejects $20M Buyout, Secures $12M Seed Funding
NanoCo, the team behind NanoClaw—a security-first alternative to OpenClaw—has secured an oversubscribed $12 million seed round following a viral launch, founders Gavriel and Lazer Cohen told TechCrunch.Valley Capital Partners led the funding round, w
Anthropic admits its AI models breached three firms during security tests
Anthropic announced Thursday that an internal probe revealed three separate incidents where its AI model, Claude, breached the systems of three organizations during cybersecurity assessments. This disclosure follows closely on the heels of OpenAI’s r





Home






