AI Threat AlertCVE-2026-19306: Langflow: chat attachment path traversal leaks server secrets
HIGHIBM Langflow's Chat Input component lets an authenticated user attach arbitrary local files to a prompt, and a path-traversal flaw in the `files` parameter of the build API lets that same user point it at absolute paths — pulling in `secret_key`, JWT signing keys, the application database, `/proc/self/environ`, and other tenants' upload directories, then shipping that content straight to the configured LLM endpoint as prompt context. This matters because it silently defeats `LANGFLOW_RESTRICT_LOCAL_FILE_ACCESS=true`, the exact control most deployments rely on to sandbox file access, meaning teams who believe they're contained are not. There's no EPSS score, no CISA KEV listing, and no public exploit or Nuclei template yet, so this isn't under active mass exploitation — but any authenticated Langflow user (a low bar in many internal deployments with shared or weak accounts) can pull JWT signing keys and pivot to full auth bypass, or read other tenants' data in multi-tenant instances. Patch to the fixed Langflow release covering CVE-2026-19306 immediately, rotate `secret_key` and JWT signing material post-patch since they may already be exposed, and until patched restrict Chat Input build requests via a reverse-proxy allowlist while auditing build-request logs for `files` parameters containing `../`, absolute paths, or references to `/proc/self/environ`.
What is the risk?
High risk despite the moderate CVSS (7.7): the confidentiality-only impact score understates the blast radius because the leaked data (secret_key, JWT signing keys) enables secondary privilege escalation and cross-tenant compromise, not just a one-time information disclosure. Exploitability is straightforward (network, low complexity, no user interaction) and the barrier is just authenticated access — low in Langflow deployments that use shared credentials, open self-signup, or weak internal auth. No public exploit, PoC, scanner template, or KEV listing exists yet, so this is not being mass-exploited today, but the technique is simple to reproduce from the vendor advisory and the value of the payoff (auth material) makes it an attractive target once details circulate.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| Langflow | pip | — | No patch |
Do you use Langflow? You're affected.
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade Langflow past 1.11.2 to the version that fixes CVE-2026-19306 as soon as IBM/Langflow publishes it. 2) Rotate
secret_keyand any JWT signing keys immediately after patching, since they may have already been exfiltrated pre-patch — treat them as burned. 3) Until patched, restrict who can reach the authenticated build API and Chat Input attachment feature (network segmentation, reverse-proxy path allowlisting, or disabling file attachments on Chat Input where not needed). 4) Audit build-request logs forfilesparameters containing traversal sequences (../), absolute paths (/etc/,/proc/,/root/), or references to other tenants' upload paths. 5) In multi-tenant deployments, verify no other tenant's upload directory was accessed by reviewing access logs against the fix window. 6) ConfirmLANGFLOW_RESTRICT_LOCAL_FILE_ACCESSenforcement covers the full file-reading surface post-patch rather than assuming the flag alone is sufficient.
What does CISA's SSVC say?
Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2026-19306?
IBM Langflow's Chat Input component lets an authenticated user attach arbitrary local files to a prompt, and a path-traversal flaw in the `files` parameter of the build API lets that same user point it at absolute paths — pulling in `secret_key`, JWT signing keys, the application database, `/proc/self/environ`, and other tenants' upload directories, then shipping that content straight to the configured LLM endpoint as prompt context. This matters because it silently defeats `LANGFLOW_RESTRICT_LOCAL_FILE_ACCESS=true`, the exact control most deployments rely on to sandbox file access, meaning teams who believe they're contained are not. There's no EPSS score, no CISA KEV listing, and no public exploit or Nuclei template yet, so this isn't under active mass exploitation — but any authenticated Langflow user (a low bar in many internal deployments with shared or weak accounts) can pull JWT signing keys and pivot to full auth bypass, or read other tenants' data in multi-tenant instances. Patch to the fixed Langflow release covering CVE-2026-19306 immediately, rotate `secret_key` and JWT signing material post-patch since they may already be exposed, and until patched restrict Chat Input build requests via a reverse-proxy allowlist while auditing build-request logs for `files` parameters containing `../`, absolute paths, or references to `/proc/self/environ`.
Is CVE-2026-19306 actively exploited?
No confirmed active exploitation of CVE-2026-19306 has been reported, but organizations should still patch proactively.
How to fix CVE-2026-19306?
1) Upgrade Langflow past 1.11.2 to the version that fixes CVE-2026-19306 as soon as IBM/Langflow publishes it. 2) Rotate `secret_key` and any JWT signing keys immediately after patching, since they may have already been exfiltrated pre-patch — treat them as burned. 3) Until patched, restrict who can reach the authenticated build API and Chat Input attachment feature (network segmentation, reverse-proxy path allowlisting, or disabling file attachments on Chat Input where not needed). 4) Audit build-request logs for `files` parameters containing traversal sequences (`../`), absolute paths (`/etc/`, `/proc/`, `/root/`), or references to other tenants' upload paths. 5) In multi-tenant deployments, verify no other tenant's upload directory was accessed by reviewing access logs against the fix window. 6) Confirm `LANGFLOW_RESTRICT_LOCAL_FILE_ACCESS` enforcement covers the full file-reading surface post-patch rather than assuming the flag alone is sufficient.
What systems are affected by CVE-2026-19306?
This vulnerability affects the following AI/ML architecture patterns: agent frameworks, LLM chat/conversation pipelines, multi-tenant SaaS AI platforms.
What is the CVSS score for CVE-2026-19306?
CVE-2026-19306 has a CVSS v3.1 base score of 7.7 (HIGH). The EPSS exploitation probability is 0.54%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0012Valid AccountsAML.T0037Data from Local SystemAML.T0086Exfiltration via AI Agent Tool InvocationCompliance Controls Affected
What are the technical details?
Original Advisory
IBM Langflow OSS 1.0.0 through 1.11.2 allows an authenticated attacker to read arbitrary files from the server filesystem — including server secret material (secret_key, JWT signing keys, the application database, /proc/self/environ, and other tenants' upload directories) — by supplying absolute paths or traversal sequences in the files parameter of an authenticated build request. The file contents were embedded as text attachments in the language model prompt and transmitted to the configured model endpoint, resulting in confidential data exfiltration. This bypassed the LANGFLOW_RESTRICT_LOCAL_FILE_ACCESS=true containment boundary, which was enforced for other file-reading components but not for the Chat Input to Message attachment pipeline.
Exploitation Scenario
An attacker who has obtained low-privilege authenticated access to a Langflow instance (e.g., via a shared internal account, a self-service signup, or a compromised low-tier credential) opens any flow that uses the Chat Input component and issues a build request with the `files` parameter set to an absolute path such as `/app/secret_key` or a traversal sequence reaching `/proc/self/environ`. Langflow embeds the referenced file's contents as a text attachment inside the message sent to the configured LLM endpoint, so the file contents flow out through the normal LLM API call rather than through a blocked file-read endpoint — evading the `LANGFLOW_RESTRICT_LOCAL_FILE_ACCESS` control that was only wired into other components. The attacker retrieves the exfiltrated secret material either directly from the model's echoed response or by intercepting/monitoring the outbound request to the LLM provider, then uses the recovered JWT signing key to forge tokens or escalate to admin-level access, or reads another tenant's uploaded documents for data theft.
Weaknesses (CWE)
CWE-22Improper Limitation of a Pathname to a Restricted DirectoryPrimaryCWE-22Improper Limitation of a Pathname to a Restricted DirectoryCWE-22 — Improper Limitation of a Pathname to a Restricted Directory: The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.
- [Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
- [Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:N/A:NReferences
Timeline
Related Vulnerabilities
CVE-2026-1056110.0Langflow: auth bypass + unauthenticated RCE (CVSS 10)
Same package: langflowCVE-2026-1013410.0Langflow: unauthenticated RCE via tool_code injection
Same package: langflowCVE-2026-333099.9langflow: Path Traversal enables file access
Same package: langflowCVE-2026-552559.9Langflow: IDOR allows cross-user flow execution
Same package: langflowCVE-2026-78739.9Langflow: authenticated RCE enables credential theft
Same package: langflow