CVE-2026-55607: Claude Code sandbox escape via Git worktree path confusion
- Vendor
- Anthropic
Claude Code versions 2.1.38 through versions before 2.1.163 contain a Git worktree path-confusion vulnerability. Worktree handling permits names corresponding to Git metadata directories and navigation to worktrees outside the sandbox context. An attacker can combine this behavior with symlink manipulation and Git fsmonitor execution during worktree operations to overwrite files in the user's home directory and execute code outside Seatbelt sandbox restrictions. Reliable exploitation requires a user to clone a malicious repository containing prompt-injection content and run Claude Code against it.
Impact, mitigation & remediation
What it means and what to do now: patch path, interim mitigations, and the assume-compromise checklist.
Impact
What an attacker gets, and what they’ve been doing with it.
Successful exploitation breaks the Claude Code sandbox boundary, allowing attacker-controlled writes to files in the user's home directory and code execution outside Seatbelt restrictions. Modification of shell initialization files can cause attacker-controlled code to execute when the affected shell initializes, potentially providing persistence with the user's privileges.
Mitigation
If you can’t patch tonight, do this now.
Until the update is installed, avoid cloning untrusted repositories and running Claude Code against them, particularly repositories containing prompt-injection content.
Remediation
Patch, then assume compromise.
Upgrade the @anthropic-ai/claude-code package to version 2.1.163 or later. Standard Claude Code auto-update delivered the fix automatically; verify that the installed version includes it.
Exploits
No public exploits tracked yet. Mallory keeps watching.
Affected products & vendors
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Frequently asked questions
- Is CVE-2026-55607 exploited in the wild?
- As of Oct 16, 2026, Mallory has no reports of CVE-2026-55607 being exploited in the wild. It is not listed in CISA KEV.
- Is there a public exploit for CVE-2026-55607?
- As of Oct 16, 2026, Mallory has found no public exploit for CVE-2026-55607.
- Is there a patch for CVE-2026-55607?
- Upgrade the @anthropic-ai/claude-code package to version 2.1.163 or later. Standard Claude Code auto-update delivered the fix automatically; verify that the installed version includes it.
- What products are affected by CVE-2026-55607?
- Anthropic Claude Code.
- How severe is CVE-2026-55607?
- CVE-2026-55607 has a CVSS v4 score of 7.7 (High) and an EPSS score of 0.69% (53rd percentile) as of Oct 16, 2026.
Recent activity
11 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.