Case · DiggingBeagle record

pgEdge Go LLM health endpoint could return provider-key fragments in error messages

A health-check error path in a Go model proxy returned upstream provider responses containing masked secret fragments to permitted health callers.

Scope

Go LLM Library proxy 0.1.0/0.1.1 with health checks and credentials configured; access depends on the application Authorize hook.

Assessment profile

Assessment method
Setting
research demonstration
Exploitation
demonstrated
Evidence
Grade B
Remediation
patch available
AI role
AGAINST AI
Basis and provenance

Errors could reveal fragments of API credentials; version 0.2.0 provides redaction, but full-key theft or downstream spending has not been shown. Classified as research_demonstration with exploit status demonstrated. Evidence grade B reflects the original paper or maintainer advisory and its stated limitations, not an independent audit of all deployments. Comparable realized harm is not established, so impact stays unrated; remediation status patch_available is limited to the cited experiment or affected software version. New source-specific boundary: Response exposure, the usability of a partial key, and verified misuse are separate outcomes; the advisory establishes a preventive rotation rationale, not observed theft.

Assessed Nov 15, 2026 using diggingbeagle.assessment/1.

At a glance

Mechanism and trust boundary

  1. 01

    Entry or starting condition

    A network caller can invoke the pgEdge Go LLM proxy health route while provider credentials are configured.

    Boundary: attacker or research precondition

  2. 02

    Crossing the relevant boundary

    The route tests upstream providers with real keys and returns provider error text without unconditional sanitization; a missing authorization hook may allow the request by default.

    Boundary: mechanism and authorization boundary

  3. 03

    Observed or reported result

    Errors could reveal fragments of API credentials; version 0.2.0 provides redaction, but full-key theft or downstream spending has not been shown.

    Boundary: measured outcome and evidentiary limit

Claims & evidence

CLM-ENR250-10The pgEdge advisory recommends reviewing and rotating API keys potentially revealed by the unauthenticated health response in addition to upgrading to the first published fixed release, 0.2.0. A code fix prevents new partial-key disclosures but cannot withdraw copies of diagnostic responses or revoke credentials already exposed.supported

Basis: reported finding

Link to claim
CLM-R245-PGEDGE-GO-HEALTH-KEYS-01The maintained advisory describes masked API-key fragments appearing in responses and calls 0.2.0 the first released fixed version, despite a conflicting pseudo-version in structured advisory metadata.supported

Basis: reported finding

Link to claim
CLM-R245-PGEDGE-GO-HEALTH-KEYS-02The /v1/health handler probes external model services with the operator's API keys and forwards provider errors without sufficient redaction; the default authorization hook does not reject requests.supported

Basis: reported finding

Link to claim
CLM-R246-PGEDGE-GO-HEALTH-KEYS-BOUNDARYThe pgEdge advisory distinguishes an unpublished fixed pseudo-version from the first released fixed tag: its maintainer says 0.2.0 is the first fixed release and no v0.1.2 release was cut, despite structured metadata carrying a pseudo-version based on a July commit.supported

Basis: reported finding

Link to claim

Unknowns and contradictions

  • A partial key fragment is not evidence of full credential theft or unauthorized inference spending. The fixed-version discrepancy must be retained until module tags and commit history are reconciled.
  • No evidence here establishes the number of production victims or a broader failure rate outside the cited evaluation.

Sources and citation

Material revision history

  1. Nov 15, 2026 · Canonical change recorded · new in release · revision 251