CVE-2026-7658: Langflow username path traversal

CISO Take

This vulnerability lets an authenticated Langflow user smuggle path traversal sequences through the username field to escape tenant containment checks, enabling deletion of arbitrary directories, destruction of other tenants' data, and removal of the platform's JWT signing key — the last of which forces a mass session invalidation for every user on the instance. It requires only low privileges and no user interaction (PR:L, UI:N), so a single compromised low-tier account is enough to trigger the impact, but the EPSS score of 0.34% (74th percentile) and CISA SSVC 'Track' rating both indicate this is not currently being weaponized in the wild — there is no public exploit, no Nuclei template, and it is not on CISA's KEV list. Langflow is a widely deployed low-code framework for building LLM agent workflows, so any multi-tenant or shared deployment (common among internal AI platform teams) carries real blast-radius risk if a single account is compromised via credential stuffing or insider misuse. Patch to the fixed Langflow release beyond 1.10.3 as the primary action, and in the interim validate/sanitize the username field server-side and monitor for anomalous directory-deletion or unexpected mass session invalidation as compensating detection.

Sources: NVD EPSS CISA SSVC ibm.com ATLAS

What is the risk?

Technical exploitability is high on paper — low attack complexity, low privileges required, no user interaction — but real-world exploitation likelihood is currently low: EPSS sits at 0.34% (74th percentile), the flaw is absent from CISA KEV, and CISA's own SSVC decision is 'Track' (lowest urgency tier), with no public exploit code or scanner template observed. The gating factor is that PR:L means an attacker needs a valid, already-authenticated account — this is not a pre-auth remote compromise. However, the impact ceiling is disproportionate to the CVSS 6.5 'medium' label in multi-tenant deployments: deleting the JWT signing key is a platform-wide availability/integrity event (all sessions invalidated), and cross-tenant data destruction breaks the core isolation guarantee a shared AI platform depends on. Treat this as low urgency for exploitation probability but high urgency for patching in any shared/multi-tenant Langflow deployment.

How does the attack unfold?

Valid account access

Attacker obtains or already controls a low-privileged, authenticated Langflow account, the only precondition for exploitation (PR:L).

AML.T0012

Path traversal injection

Attacker sets the username field to a value containing path traversal sequences, which Langflow fails to validate or sanitize (CWE-22).

Containment bypass

The unsanitized username is used to construct file system paths for tenant-scoped operations, letting the traversal escape the intended tenant directory and reach arbitrary paths, including other tenants' data and the JWT signing key store.

Data destruction and session invalidation

Attacker deletes arbitrary directories, destroying cross-tenant data and/or removing the JWT signing key, which invalidates every active session platform-wide.

Valid account access

Attacker obtains or already controls a low-privileged, authenticated Langflow account, the only precondition for exploitation (PR:L).

AML.T0012

Path traversal injection

Attacker sets the username field to a value containing path traversal sequences, which Langflow fails to validate or sanitize (CWE-22).

Containment bypass

The unsanitized username is used to construct file system paths for tenant-scoped operations, letting the traversal escape the intended tenant directory and reach arbitrary paths, including other tenants' data and the JWT signing key store.

Data destruction and session invalidation

Attacker deletes arbitrary directories, destroying cross-tenant data and/or removing the JWT signing key, which invalidates every active session platform-wide.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Langflow	pip	—	No patch
155.0K Pushed 2d ago 17% patched ~70d to patch Full package profile →

Do you use Langflow? You're affected.

How severe is it?

CVSS 3.1

6.5 / 10

EPSS

0.3%

chance of exploitation in 30 days

Higher than 27% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Moderate

What is the attack surface?

AV Network

AC Low

PR Low

UI None

S Unchanged

C None

I High

A None

What should I do?

1 step

Upgrade Langflow past 1.10.3 to the IBM-patched release referenced in the vendor advisory (support.ibm.com, node/7282647) as soon as it is validated in a test environment. Until patched, add server-side validation/allow-listing on the username field to reject path traversal sequences (../, absolute paths, encoded variants) rather than relying on client-side checks. Run the Langflow process under a filesystem user with least-privilege write/delete access so traversal cannot reach directories outside its intended scope, as defense in depth. Restrict who can create or self-register low-privileged accounts, since PR:L is the only precondition for exploitation. Monitor for unexpected directory deletions in tenant data paths and for unplanned mass session invalidation events (JWT key rotation/deletion), both of which are strong indicators of exploitation attempts.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Auth Bypass DoS Framework Agent AML.T0012 - Valid Accounts

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6 - System security

NIST AI RMF

MEASURE 2.7 - AI system security and resilience is evaluated and documented

Frequently Asked Questions

What is CVE-2026-7658?

This vulnerability lets an authenticated Langflow user smuggle path traversal sequences through the username field to escape tenant containment checks, enabling deletion of arbitrary directories, destruction of other tenants' data, and removal of the platform's JWT signing key — the last of which forces a mass session invalidation for every user on the instance. It requires only low privileges and no user interaction (PR:L, UI:N), so a single compromised low-tier account is enough to trigger the impact, but the EPSS score of 0.34% (74th percentile) and CISA SSVC 'Track' rating both indicate this is not currently being weaponized in the wild — there is no public exploit, no Nuclei template, and it is not on CISA's KEV list. Langflow is a widely deployed low-code framework for building LLM agent workflows, so any multi-tenant or shared deployment (common among internal AI platform teams) carries real blast-radius risk if a single account is compromised via credential stuffing or insider misuse. Patch to the fixed Langflow release beyond 1.10.3 as the primary action, and in the interim validate/sanitize the username field server-side and monitor for anomalous directory-deletion or unexpected mass session invalidation as compensating detection.

Is CVE-2026-7658 actively exploited?

No confirmed active exploitation of CVE-2026-7658 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-7658?

Upgrade Langflow past 1.10.3 to the IBM-patched release referenced in the vendor advisory (support.ibm.com, node/7282647) as soon as it is validated in a test environment. Until patched, add server-side validation/allow-listing on the username field to reject path traversal sequences (../, absolute paths, encoded variants) rather than relying on client-side checks. Run the Langflow process under a filesystem user with least-privilege write/delete access so traversal cannot reach directories outside its intended scope, as defense in depth. Restrict who can create or self-register low-privileged accounts, since PR:L is the only precondition for exploitation. Monitor for unexpected directory deletions in tenant data paths and for unplanned mass session invalidation events (JWT key rotation/deletion), both of which are strong indicators of exploitation attempts.

What systems are affected by CVE-2026-7658?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, multi-tenant AI platforms, LLM workflow orchestration.

What is the CVSS score for CVE-2026-7658?

CVE-2026-7658 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.34%.

What is the AI security impact?

Affected AI Architectures

agent frameworksmulti-tenant AI platformsLLM workflow orchestration

MITRE ATLAS Techniques

AML.T0012 Valid Accounts

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: A.6.2.6

NIST AI RMF: MEASURE 2.7

What are the technical details?

Original Advisory

IBM Langflow OSS 1.0.0 through 1.10.3 does not properly validate the username field, allowing attackers to inject path traversal sequences and bypass containment checks. This enables multiple severe impacts, including arbitrary directory deletion, cross-tenant data destruction, and JWT signing key deletion leading to session invalidation.

Exploitation Scenario

An attacker who holds or has stolen a low-privileged Langflow account (e.g., via credential stuffing, an insider, or a phished low-tier user) crafts a username containing path traversal sequences such as '../../..' during account creation or profile update. Because Langflow does not validate or normalize this field server-side, the value later gets used to construct file system paths for tenant-scoped file operations. When those operations execute, the traversal sequences escape the intended tenant sandbox, allowing the attacker to delete arbitrary directories elsewhere on disk — including other tenants' flow/data directories or the directory holding the JWT signing key. Deleting the signing key immediately invalidates every active session platform-wide, forcing all users to re-authenticate and creating operational chaos that can mask the attacker's initial access or further activity.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') Primary CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): 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.