CVE-2026-33497
highpriority · track · 37EPSS 0.196 · 97 %ileDEEUUSEPSS · scored 2026-10-27
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.7.1, in the download_profile_picture function of the /profile_pictures/{folder_name}/{file_name} endpoint, the folder_name and file_name parameters are not strictly filtered, which allows the secret_key to be read across directories. Version 1.7.1 contains a patch.
cvss
Technical severity, 0–10. One row per scoring source.
base score 0–10 · higher = more severe · hover a vector to decode
ssvc
A stakeholder decision — act, attend, or track — from exploitation and impact.
weaknesses
The CWE weakness classes this flaw belongs to.
affected products (5)
Products and version ranges known to be affected.
advisories (3)
Vendor and ecosystem advisories describing this.
- WID-SEC-W-2026-0823Langflow: Mehrere SchwachstellenDeutsch2026-03-22csaf · cert_bundDE
Ein Angreifer kann mehrere Schwachstellen in Langflow ausnutzen, um beliebigen Programmcode auszuführen, um Informationen offenzulegen, und potenziell um Daten zu manipulieren.
- GHSA-ph9w-r52h-28p7langflow: /profile_pictures/{folder_name}/{file_name} endpoint file reading2026-03-20ghsa · osvUS
## Vulnerability ### Path Traversal in `GET /api/v1/files/profile_pictures/{folder_name}/{file_name}` The `download_profile_picture` function in `src/backend/base/langflow/api/v1/files.py` constructed file paths by directly concatenating the user-supplied `folder_name` and `file_name` path parameters without sanitization or boundary validation. The resulting path was passed to the filesystem without verifying it remained within the intended directory. An unauthenticated attacker could supply traversal sequences (e.g. `../secret_key`) to navigate outside the profile pictures directory and read arbitrary files on the server filesystem. This exposed the server to: - **Sensitive file disclosure** — any file readable by the application process could be retrieved - **Secret key exfiltration** — the application's `secret_key` file, used as JWT signing material, could be read directly via `../secret_key` - **Authentication bypass** — with the `secret_key` in hand, an attacker can forge valid JWT tokens and authenticate as any user, including administrators --- ## Proof of Concept ```bash curl --path-as-is 'http://<host>:7860/api/v1/files/profile_pictures/../secret_key' ``` A successful response returns the raw secret key value used to sign all JWT authentication tokens in the instance. --- ## Fix The fix was applied in `src/backend/base/langflow/api/v1/files.py` (PR #12263). Two layers of defense were introduced: **1. Typed path validation** — the `folder_name` and `file_name` parameters were changed from plain `str` to `ValidatedFolderName` and `ValidatedFileName` annotated types that reject traversal characters at the FastAPI input layer. **2. Path containment check** — `Path.name` is used to strip any directory component from the inputs before path construction, and `Path.is_relative_to()` verifies the resolved path remains within the allowed base directory. This replaces the previous `startswith()` check, which was susceptible to prefix-ambiguity bugs. ```diff @router.get("/profile_pictures/{folder_name}/{file_name}") async def download_profile_picture( - folder_name: str, - file_name: str, + folder_name: ValidatedFolderName, + file_name: ValidatedFileName, settings_service: Annotated[SettingsService, Depends(get_settings_service)], ): ``` ```diff - file_path = (config_path / "profile_pictures" / folder_name / file_name).resolve() + safe_folder = Path(folder_name).name + safe_file = Path(file_name).name + file_path = (config_path / "profile_pictures" / safe_folder / safe_file).resolve() allowed_base = (config_path / "profile_pictures").resolve() - if not str(file_path).startswith(str(allowed_base)): - raise HTTPException(status_code=404, detail="Profile picture not found") + if not file_path.is_relative_to(allowed_base): + raise HTTPException(status_code=404, detail="Profile picture not found") ``` --- ## Workarounds If you cannot upgrade immediately, restrict network access to the `/api/v1/files/profile_pictures/` endpoint at the reverse-proxy or firewall level. Rotating the `secret_key` is strongly recommended if exposure cannot be ruled out. --- ## Acknowledgements We thank the security researcher who responsibly disclosed this vulnerability. - [r00tuser111](https://github.com/r00tuser111)
- PYSEC-2026-812026-03-24vendor · osvUS
Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.7.1, in the download_profile_picture function of the /profile_pictures/{folder_name}/{file_name} endpoint, the folder_name and file_name parameters are not strictly filtered, which allows the secret_key to be read across directories. Version 1.7.1 contains a patch.
also known as (6)
Other names and identifiers for this entity.
sources
Which sources assert this record, and when.
references (4)
External citations — patches, advisories, analysis.
- https://github.com/langflow-ai/langflow/releasesPatchcnnvd
- https://github.com/langflow-ai/langflow/security/advisories/GHSA-ph9w-r52h-28p7cnnvd
- https://github.com/langflow-ai/langflow/security/advisories/GHSA-ph9w-r52h-28p7enisa_euvd
- https://github.com/langflow-ai/langflow/security/advisories/GHSA-ph9w-r52h-28p7ExploitVendor Advisorynvd
Attackers weaponise vulnerabilities like this against internet-exposed services. See what is exposed on your perimeter with a free passive scan.
Free scan