Cyber Resilience

CVE-2025-64504

Langfuse 2.70.0 – 2.95.11

Published
10 November 2025
Modified
02 December 2025
Patch / advisory
CVSS Score v3.1 5.0
Click a component to see what it means
Raw vectorCVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:L/I:N/A:N
EPSS Score 0.0032 25th percentile
Risk Priority 38 floored blend · peak EPSS

Summary

CVE-2025-64504 is a medium-severity Exposure of Sensitive Information Through Data Queries (CWE-202) vulnerability in Langfuse Langfuse. Its CVSS base score is 5.0 (Medium).

Operationally, exploitation aligns with the MITRE ATT&CK technique Data from Information Repositories (T1213); ranked at the 25th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog.

This vulnerability is AI-related — categorised as LLM Application Platforms; in the Privacy and Disclosure risk domain.

The strongest mitigations our analysis identified map to AC-23 (Data Mining Protection) — see the control section below for these in your framework.

EU & UK References

Vulnerability Data

Langfuse is an open source large language model engineering platform. Starting in version 2.70.0 and prior to versions 2.95.11 and 3.124.1, in certain project membership APIs, the server trusted a user‑controlled orgId and used it in authorization checks. As a…

more

result, any authenticated user on the same Langfuse instance could enumerate names and email addresses of users in another organization if they knew the target organization’s ID. Disclosure is limited to names and email addresses of members/invitees. No customer data such as traces, prompts, or evaluations is exposed or accessible. For Langfuse Cloud, the maintainers ran a thorough investigation of access logs of the last 30 days and could not find any evidence that this vulnerability was exploited. For most self-hosting deployments, the attack surface is significantly reduced given an SSO provider is configured and email/password sign-up is disabled. In these cases, only users who authenticate via the Enterprise SSO IdP (e.g. Okta) would be able to exploit this vulnerability to access the member list, i.e. internal users getting access to a list of other internal users. In order to exploit the vulnerability, the actor must have a valid Langfuse user account within the same instance, know the target orgId, and use the request made to the API that powers the frontend membership tables, including their project/user authentication token, while changing the orgId to the target organization. Langfuse Cloud (EU, US, HIPAA) were affected until fix deployment on November 1, 2025. The maintainers reviewed the Langfuse Cloud access logs from the past 30 days and found no evidence that this vulnerability was exploited. Self-Hosted versions which contain patches include v2.95.11 for major version 2 and v3.124.1 for major version 3. There are no known workarounds. Upgrading is required to fully mitigate this issue.

CWE(s)

AI Security AnalysisAI

AI Category
LLM Application Platforms
Risk Domain
Privacy and Disclosure
OWASP Top 10 for LLMs 2025
None mapped
Classification Reason
Matched keywords: langfuse, large language model

Related Threats

MITRE ATT&CK Enterprise Techniques

T1213 Data from Information Repositories Collection
Adversaries may leverage information repositories to mine valuable information.
T1213.006 Databases Collection
Adversaries may leverage databases to mine valuable information.
Derived from this CVE’s CWE(s) via the direct CWE→ATT&CK cross-walk.

CVEs Like This One

CVE-2026-24055Same product: Langfuse Langfuse
CVE-2026-41487Same product: Langfuse Langfuse
CVE-2025-65107Same product: Langfuse Langfuse
CVE-2025-59305Same product: Langfuse Langfuse
CVE-2025-9799Same product: Langfuse Langfuse
CVE-2024-2088Shared CWE-202
CVE-2024-20388Shared CWE-202
CVE-2025-29981Shared CWE-202
CVE-2026-33530Shared CWE-202
CVE-2023-20215Shared CWE-202

Affected Assets

langfuse
langfuse
2.70.0 — 2.95.11 · 3.0.0 — 3.124.1

Mitigating Controls

Mitigating Controls (NIST 800-53 r5) AI

AC-23 directly requires mechanisms to protect against unauthorized data mining and inference from statistical queries that would expose sensitive information.

Mitigating Controls (NIST CSF 2.0) AI

Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.

PR.AA-05 mostly match
prevents

Least-privilege query permissions directly limit the data an attacker can request or infer.

DE.CM-03 partial match
prevents

Behavior analytics on query activity can detect inference attempts but does not prevent exposure at query time.

PR.DS-10 partial match
prevents

Protecting data-in-use reduces what remains available for inference via queries.

Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI

Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.

prevents

Access control limits who can run queries that could expose sensitive information via inference.

prevents

Granular access rights reduce the ability of users to craft inference queries.

prevents

Data masking prevents inference by obscuring sensitive values returned in query results.

prevents

Information access restriction directly limits query scope that could lead to inference.

mitigates

Classification helps identify sensitive data that must be protected from inference attacks.

finds

DLP can detect and block queries or result sets that risk exposing sensitive information.

References