Finding and attribution
In non-sandbox-contained operation, apply_patch path resolution can accept absolute or traversal paths and write or delete files outside the configured workspace. The AI SAST engine identified the code path; the researcher separately demonstrated that fabricated conversation history can bypass LLM guardrails and induce the tool call. This requires the model to cooperate and is not a generic direct HTTP path traversal.
Product: OpenClaw. AI system: Endor Labs AI SAST engine. Team / organization: Endor Labs.
CVE identifiers: Not established in the cited evidence.
AI and human contribution (source account): Endor Labs’ case-specific article states that its AI SAST engine traced attacker-controlled patch text through path resolution to filesystem write/delete sinks; researcher Peyton Kennedy then validated the exploit and reports a separate LLM guardrail-bypass path. The LLM cooperation is a condition of the demonstrated attack, not the AI SAST discovery method. The article was updated after initial publication, so the exact first public date of its current attribution text is not established.
Human involvement is only described where the source account establishes it; unknown involvement does not mean none. Vendor confirmation of a flaw does not confirm every claim about AI contribution.
Discovery context: The upstream security advisory documents the public disclosure and fix. The sources do not establish whether defenders knew of the issue or whether it was exploited before disclosure.
Remediation status: fix documented. OpenClaw 2026.2.14 and later are listed as patched. Safe default workspaceOnly=true confines paths; maintainers link PR #16405.
A source-documented fix does not establish installed patches or current safety. Source checking does not independently reproduce the finding or establish observed exploitation.
Disclosed 19 Feb 2026. AI attribution date not established by the source. Sources checked 16 Nov 2026; these are separate dates, not a live installation check.
Recorded CVSS: Not available · no assessment established here. Documented technical effects: data modification.
Severity, consequences and exploitation evidence
Case grouping uses the highest recorded CVSS score, with its version-specific severity band. It is descriptive, not a combined or editorial risk index. All recorded assessments remain below; scores from different versions and sources may disagree. CVSS specifications define version-specific scoring and severity bands.
No source-backed CVSS assessment is established in this catalog.
Case-level evidence: data modification.
Prerequisites / affected scope: Non-sandboxed path resolution / absent sandboxRoot or explicit workspaceOnly:false; apply_patch must be enabled and an attacker must be able to trigger LLM-mediated tool execution..
Exploitation evidence: The sources report validation or describe attack conditions but do not report observed exploitation.. Impact source · checked 16 Nov 2026.
Documented fix: OpenClaw 2026.2.14 and later are listed as patched. Safe default workspaceOnly=true confines paths; maintainers link PR #16405.