AEV-2026-0017
Codex CLI's own Git metadata reads honoured a repository's core.fsmonitor before a fix shipped (CVE-2026-19592)
Codex CLI and Codex Desktop collected Git repository metadata without disabling a repository's own core.fsmonitor setting, letting an attacker named helper run with the user's privileges outside Codex's own command sandbox. OpenAI's own merged fixes predate the September 2026 CVE by months; Manifold Security's later report of the same class was closed as a duplicate of the earlier, independently filed finding.
Affected
- Organisation
- OpenAI
- Product
- Codex CLI and Codex Desktop
- Component
- Automatic Git repository metadata collection (Git status and diff subprocesses)
- Versions
- Codex CLI for Windows, macOS and Linux and Codex Desktop for Windows and macOS, before the fixes merged in pull requests 22652 and 26880; CVE-2026-19592 published 1 September 2026
- Configurations
- Repository's .git/config sets an executable core.fsmonitor helper
Execution authority facts
Fields the evidence does not establish are shown as unknown rather than guessed.
- Intended task
- Codex's own automatic collection of Git repository metadata for context
- Agent
- Codex CLI and Codex Desktop's internal Git subprocess handling
- Delegating actor
- The developer opening or using the repository with Codex
- Action
- Executed a repository named core.fsmonitor helper outside Codex's own command sandbox and without a user approval prompt
- Target resource
- The host running Codex, with the user's own privileges
- Environment
- A local developer workstation or desktop session running Codex CLI or Codex Desktop against an untrusted repository
- Credentials used
- unknown
- Privileges available
- The user's own account privileges, able to read, change or delete the user's files and reach other resources available to that account
- Authority presented
- None; the action ran as internal Git metadata collection rather than a tool call the model or user approved
- Authority required
- Authorization for host code execution with the user's own privileges, outside Codex's own command sandbox
- Applicable policy
- Codex's own command sandbox and tool approval model, which this action ran outside of
- Approval mechanism
- none
- Required approver
- unknown
- Independent approval
- no
- Action binding
- unknown
- Sequence context
- Ran during Codex's own automatic Git metadata collection, before any model directed tool call
Impact
- Consequence
- Attacker controlled code able to read, change or delete the user's files and reach other resources available to the user's account, per the CVE record's own description
- Reach
- single
- Reversibility
- unknown
- Detectability
- silent
- Propagation
- potential
- Recovery
- unknown
Evidence
Primary sources
- [codex] Ignore fsmonitor config in Git metadata reads (Pull Request 22652) (GitHub, openai/codex)
- [codex] preserve fsmonitor for worktree Git reads (Pull Request 26880) (GitHub, openai/codex)
Supporting sources
- CVE-2026-19592 record, CVE Program (CVE Program)
- GitSpawn: A Single Flaw Lets Untrusted Repos Run Code in Claude Code, Codex, Cursor, and Grok (Manifold Security)
- Codex CLI and Codex Desktop automatically collected Git repository metadata without disabling core.fsmonitor (GHSA-26wp-42v3-96xp) (GitHub Advisory Database)
- Reproduction status
- OpenAI's own merged pull requests and their regression handling verified directly; not independently reproduced by Moona Intelligence against a running Codex instance.
- Evidence state
- Confirmed
Known unknowns
- The exact Codex CLI and Desktop version numbers CVE-2026-19592's own record considers affected and fixed.
- Whether Manifold's own report, closed as a duplicate, targeted exactly the mechanism these two pull requests close or a further variant.
Limitations
- CVE-2026-19592's own record at cve.org could not be directly fetched in this session and is corroborated through search. The two OpenAI pull requests cited as primary evidence were fetched directly and are the strongest basis for this entry's mechanism and timeline.
Claim provenance
- verified
OpenAI's own merged pull request, read directly, stops Codex's internal Git metadata and status subprocesses from honouring a repository's own core.fsmonitor setting, merged 14 May 2026 and referencing an internal identifier (ZDI_31270) consistent with a report through the Zero Day Initiative predating Manifold's own disclosure.
The Agent Ran Code Before It Asked Whether You Trusted the Repository - verified
OpenAI's own merged follow up pull request, read directly, preserves Git's built in file system monitor daemon only when it can be verified and otherwise disables core.fsmonitor, merged 9 June 2026 to close a performance regression the first fix introduced while still keeping a repository from selecting an executable helper of its own.
The Agent Ran Code Before It Asked Whether You Trusted the Repository - verified
GitHub's own advisory for CVE-2026-19592, read directly, confirms CWE-15 and a CVSS 3.1 base score of 7.3, states the helper runs outside Codex's own command sandbox and without a user approval prompt, and names pull request openai/codex#22652 as its reference, corroborating this entry's existing primary evidence rather than adding a new claim.
The Agent Ran Code Before It Asked Whether You Trusted the Repository
