Anthropic Silently Patches Claude Code Sandbox Bypass

Anthropic has silently patched a sandbox bypass vulnerability in Claude Code. The researcher who discovered the flaw noted it could be chained with a prompt injection attack to exfiltrate data from the host environment. No public CVE or security advisory was issued alongside the fix.

The vulnerability is particularly relevant for users running Claude Code in automated pipelines, CI/CD systems, or agentic workflows where untrusted input may reach the model. Users should ensure they are running the latest version of Claude Code and review any agentic workflows that process externally supplied content for prompt injection exposure.