CVE-2024-10272: Lunary Auth Bypass Vulnerability

CVE-2024-10272 Overview

CVE-2024-10272 is a broken access control vulnerability in lunary-ai/lunary, an open-source platform for large language model (LLM) observability and analytics. The vulnerability allows unauthenticated attackers to retrieve dataset contents by sending a GET request to the /v1/datasets endpoint without a valid authorization token. The flaw is categorized under [CWE-862] Missing Authorization and affects the latest release at the time of disclosure.

Critical Impact
Unauthenticated remote attackers can access and exfiltrate the contents of any dataset stored in Lunary, exposing training data, prompts, and other sensitive LLM assets.

Affected Products

Lunary AI Lunary (all versions prior to the fix in commit 35dd4af)
The /v1/datasets API endpoint
Self-hosted deployments of lunary-ai/lunary

Discovery Timeline

2025-03-20 - CVE-2024-10272 published to NVD
2026-06-17 - Last updated in NVD database

Technical Details for CVE-2024-10272

Vulnerability Analysis

The vulnerability resides in the dataset retrieval logic of the Lunary API. The /v1/datasets route handler processes incoming GET requests without validating the presence or authenticity of an authorization token. As a result, any network-reachable client can enumerate and read dataset content that should be gated behind authenticated sessions.

The issue is classified as [CWE-862] Missing Authorization. Lunary is used to store prompts, evaluation data, and model interaction records tied to LLM applications. Unauthorized access to these datasets can leak proprietary prompts, personal data submitted by end users, and internal evaluation artifacts.

The EPSS score for this vulnerability is 0.561%, placing it in the 42nd percentile for likelihood of exploitation. No public proof-of-concept exploit or CISA KEV listing has been recorded.

Root Cause

The root cause is a missing authorization check on the /v1/datasets endpoint. The handler serves dataset records to any requester regardless of authentication state. Access control middleware was either not applied to the route or was bypassed by the request handler logic. The upstream fix in commit 35dd4af0001a54ccb14276a1546eb977f82c0c5e adds the required authorization enforcement.

Attack Vector

Exploitation requires only network access to a Lunary instance. An attacker sends an unauthenticated HTTP GET request to /v1/datasets and receives dataset data in the response. No credentials, user interaction, or elevated privileges are required. The attack is trivial to automate against internet-exposed Lunary deployments.

Refer to the Huntr Bounty Listing for the disclosure details and the GitHub Commit Change for the patch.

Detection Methods for CVE-2024-10272

Indicators of Compromise

Unauthenticated HTTP GET requests to /v1/datasets returning HTTP 200 responses with dataset content
Access log entries for /v1/datasets lacking an Authorization header or bearer token
Anomalous outbound data volume from the Lunary application server following dataset endpoint access
Requests to /v1/datasets originating from unexpected IP ranges or automated user agents

Detection Strategies

Review web server and reverse proxy logs for GET requests to /v1/datasets where the Authorization header is absent or malformed
Correlate 200-status responses on dataset endpoints with the presence of authentication tokens
Deploy a web application firewall (WAF) rule to flag unauthenticated access to /v1/* API paths

Monitoring Recommendations

Enable verbose API access logging on all Lunary endpoints, capturing headers and response sizes
Alert on repeated requests to dataset routes from a single source within a short time window
Baseline normal API consumption patterns and alert on deviations that could indicate scraping activity

How to Mitigate CVE-2024-10272

Immediate Actions Required

Upgrade Lunary to a version containing commit 35dd4af0001a54ccb14276a1546eb977f82c0c5e or later
Restrict network access to the Lunary API by placing it behind a VPN, reverse proxy with authentication, or IP allowlist until patching is complete
Audit access logs for prior unauthenticated requests to /v1/datasets and assume exposure of any datasets returned
Rotate any secrets, prompts, or credentials that may have been stored in exposed datasets

Patch Information

The fix is available in the GitHub Commit Change, which adds the missing authorization enforcement on the datasets endpoint. Operators of self-hosted Lunary instances should pull the latest release from the official repository and redeploy. Users of managed Lunary services should confirm with the vendor that the patched version is in production.

Workarounds

Place the Lunary API behind an authenticating reverse proxy that rejects requests missing a valid Authorization header
Block external access to /v1/datasets at the network edge until the patch is applied
Restrict the Lunary service to internal networks only, preventing internet-facing exposure

bash

# Example nginx configuration to require an Authorization header on /v1/datasets
location /v1/datasets {
    if ($http_authorization = "") {
        return 401;
    }
    proxy_pass http://lunary_backend;
}