CVE-2026-75858: Silent Remote Code Execution via Approval Bypass in CodeWhale Interactive Tools
Vulnerability ID: CVE-2026-75858
CVSS Score: 7.8
Published: 2026-09-04
CVE-2026-75858 is a critical authorization bypass vulnerability in CodeWhale's interactive execution tools, allowing silent, unprompted execution of model-supplied Python and shell commands on the host machine. The defect affects versions between 0.8.41 and 0.8.64, bypassing any configured approval policies via indirect prompt injection.
TL;DR
A logical flaw in CodeWhale's tool engine allows local and remote code execution by bypassing the '--approval-policy' via 'rlm_eval' and 'exec_shell_interact' tools.
⚠️ Exploit Status: POC
Technical Details
- CWE ID: CWE-94
- Attack Vector: Local (User-assisted)
- CVSS v3.1 Score: 7.8 (High)
- CVSS v4.0 Score: 8.5 (High)
- EPSS Score: 0.00267 (18.59th percentile)
- Exploit Status: PoC available
- Impact: Arbitrary Code Execution
Affected Systems
- CodeWhale AI Agent Workspace
- codewhale-tui Rust Crate
- codewhale Rust Crate
-
codewhale-tui: >= 0.8.41, < 0.8.64 (Fixed in:
0.8.64) -
codewhale: >= 0.8.41, < 0.8.64 (Fixed in:
0.8.64)
Code Analysis
Commit: 57f3c89
Fixes approval requirement logic by forcing RlmEvalTool and ShellInteractTool to require explicit approval.
Mitigation Strategies
- Upgrade codewhale and codewhale-tui to version 0.8.64 or higher.
- Sandbox execution workspaces inside isolated Docker containers.
- Restrict the agent's outbound network connections to external domains.
Remediation Steps:
- Run 'cargo update -p codewhale-tui' to pull the patched version (>= 0.8.64).
- Restart any running instances of the CodeWhale terminal user interface (TUI).
- Implement endpoint process auditing policies to track child processes of the CodeWhale agent.
References
Read the full report for CVE-2026-75858 on our website for more details including interactive diagrams and full exploit analysis.
Top comments (0)