Anthropic has tightened security for its Claude Code agent by disabling sandbox auto-approval by default.
Anthropic has updated the default security settings for its terminal-based AI coding agent, Claude Code, now requiring users to explicitly enable sandbox auto-approvals. This change addresses potential safety concerns by ensuring that potentially risky shell commands and file system operations within the sandbox are subject to user review, rather than being automatically approved by default.
Anthropic is prioritizing security over ease-of-use for their agentic coding tool, which is a sensible move given the inherent risks of autonomous agents with file system access. While this adds friction to the developer workflow, it mitigates the risk of accidental or malicious code execution.
- –Security-first approach: By defaulting to safer settings, Anthropic protects users from running unvetted code.
- –Increased friction: Users now need to manually configure their environment, which may slow down adoption for those seeking a "set-and-forget" experience.
- –Encourages safer practices: This change nudges developers towards understanding the security implications of running AI agents in their terminal.
DISCOVERED
21m ago
2026-11-17
PUBLISHED
26m ago
2026-11-17
RELEVANCE
AUTHOR
AdamHoltererer