CVE-2026-125328: pgEdge Go LLM Library credential disclosure in error messages

CVE-2026-125328 · Severity: medium · CVSS 5.8 · Published 2026-11-10

pgEdge's Go LLM library provides a unified interface for calling multiple AI/LLM providers like OpenAI and others. Due to missing access controls and credential leakage in error messages, an unauthenticated attacker can query the health endpoint and obtain fragments of the operator's third-party API keys from error responses, potentially enabling unauthorized API access.

Vendors: Go, pgEdge.

Quick answers

Is CVE-2026-125328 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-125328 being exploited in the wild.
How severe is CVE-2026-125328?
Medium severity, with a CVSS score of 5.8, published 10 November 2026.
Which products and versions are affected?
github.com/pgEdge/pgedge-go-llm-lib (Go) < 0.1.2-0.20260729024727-7531a55d368b; pgEdge Go LLM Library 0.1.0 to before 0.2.0.
Is there a fix for CVE-2026-125328?
Yes: fixed in 0.2.0. Until then: Configure the Authorize hook in llm/proxy/proxy.go to restrict access to the proxy; do not rely on the default-allow behavior. Additionally, ensure provider error messages are redacted before being returned to untrusted clients. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-125328 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-125328?
Check whether you run github.com/pgEdge/pgedge-go-llm-lib (Go) or pgEdge Go LLM Library at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • Fixed in: 0.2.0

Configure the Authorize hook in llm/proxy/proxy.go to restrict access to the proxy; do not rely on the default-allow behavior. Additionally, ensure provider error messages are redacted before being returned to untrusted clients.

Technical details

The vulnerability stems from two issues: the GET /v1/health endpoint in llm/proxy/handlers.go uses the operator's real provider credentials to ping services and returns unredacted error text, and the Authorize hook in llm/proxy/proxy.go defaults to permitting access when not configured. When a provider rejects the credentials, error messages quote back a partially masked form of the submitted API key. An unauthenticated remote attacker can exploit both the default-allow authorization and credential leakage to extract API key fragments.

Affected products

  • pgEdge Go LLM Library 0.1.0 to before 0.2.0

Identifiers

  • CVE-2026-125328
  • GHSA-8835-xmqj-354h

Timeline

  • 2026-11-10: disclosed

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.