CVE-2025-3248
Langflow before 1.3.0: unauthenticated remote code execution via POST /api/v1/validate/code
- Severity
- critical
- Affected product
- Langflow
- Affected versions
- Langflow < 1.3.0
- Fixed in
- Langflow 1.3.0
- Added to NewScan
- 2026-07-23
- Detected by
- NewScan — free, self-hosted
How NewScan reports it
COMPONENT VERSION RANGE
NewScan fingerprints Langflow from its response and reports this CVE when the detected version falls inside the affected range below.
CISA-flagged as actively exploited (surfaced by the 2026-08-05 research feed, not the NVD window). The endpoint compiles and evaluates attacker-supplied Python with no authentication at all - it is reachable whether or not the instance uses LANGFLOW_AUTO_LOGIN, which is what separates this from the auto-login exposure the interfaces row covers. CONFIRMED LIVE 2026-08-05 on the vulhub env (cvectl slug langflow-rce, langflowai/langflow:1.2.0): POST /api/v1/validate/code answered 200 {"imports":{"errors":[]},"function":{"errors":[]}} with no credential. Written as a version-gated advisory rather than an active probe on purpose: proving it in-band means actually executing code on someone's host. Version source is /api/v1/version ({"version":"1.2.0",...,"package":"Langflow"}); that path is NOT in the discovery wordlist, so this row only becomes reachable with the interface_tools change shipped alongside it - i.e. it is live at the next image release, not from the pack bundle.
References
Scan for this yourself — local, in-band scanning is free.
Get NewScan (FREE) →