CVE-2026-83816
CodeWhale’s read-only classifier could silently authorize executable commands
codewhale-hq · CodeWhale coding agent
First documented disclosure · Review checkpoint
Summary
Commands classified as read-only could launch another program without the expected approval prompt.
Severity and exploitation evidence
High · 8.1/10 · CVSS 3.1. Severity source.
Exploitation not established: Malicious exploitation status remains unverified in the reviewed sources.
Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.
How the weakness works
The classifier ignored utility options that changed a search command into program execution.
Conditions required
Attacker-controlled content influences an agent tool call in the default interactive profile.
Potential impact
Commands can execute on the developer host and access or change its data.
Affected versions and configurations
Before 0.9.0.
Fix and mitigation status
0.9.0.
Practical next steps
Upgrade to 0.9.0 or later and review agent-executed commands on affected hosts.
Operating lesson
Evaluate the behavior of command options before granting automatic approval.
Evidence and limits
Original disclosure 2026-11-12. The CVE Program published this record 2026-11-16.
Sources
- GitHub_M CVE authority record
Primary source ·
- CodeWhale coding agent original advisory
Primary source ·
- ENISA EUVD-2026-126598
Primary source ·