Case · DiggingBeagle record

A caller header could replace ContextForge managed upstream credentials

IBM ContextForge accepted an authenticated caller but allowed that caller to influence which upstream credential the gateway used for downstream MCP operations.

Public disclosure: Nov 13, 2026

Scope

An authenticated gateway user can supply an upstream-authorization header to a gateway serving downstream tools.

UnratedAI role: AGAINST AIAssessment method

At a glance

Mechanism and trust boundary

  1. 01

    Starting condition

    An authenticated gateway user can supply an upstream-authorization header to a gateway serving downstream tools.

    Boundary: precondition and attacker capability

  2. 02

    Trust boundary crossed or evaluated

    The X-Upstream-Authorization header could override managed OAuth authorization-code or client-credentials tokens, bearer tokens and Basic credentials. Authentication at the gateway did not guarantee that a request reached the upstream server under the principal authorized by gateway policy.

    Boundary: mechanism

  3. 03

    Observed consequence

    The effective upstream identity, rather than the gateway login, determines the rights and audit attribution at the downstream API.

    Boundary: observed effect and limitations

Claims & evidence

CLM-NOV15-CF110279-MECHIBM ContextForge accepted an authenticated caller but allowed that caller to influence which upstream credential the gateway used for downstream MCP operations. The X-Upstream-Authorization header could override managed OAuth authorization-code or client-credentials tokens, bearer tokens and Basic credentials. Authentication at the gateway did not guarantee that a request reached the upstream server under the principal authorized by gateway policy.supported

Basis: reported finding

Link to claim
CLM-NOV15-CF110279-SCOPECVE-2026-110279 affects the stated 1.0.0 through 1.1.0 range and was addressed in 1.1.1. The vendor describes a possible permission and audit-attribution bypass; it has not supplied a verified in-the-wild victim or stolen-key census. The published evidence is a security advisory and patch, not a demonstrated production compromise. One-time authentication and explicitly unmanaged credentials are separate intended flows and should not be treated as identical to the managed-credential path.supported

Basis: reported finding

Link to claim

Implications

The effective upstream identity, rather than the gateway login, determines the rights and audit attribution at the downstream API.

Controls and mitigations

  • Upgrade managed-credential gateways to a fixed release and reject caller-supplied upstream-credential overrides unless an explicit policy grants that mechanism.
  • Test low-rights and high-rights principals against the same tool and compare both upstream action and audit identity.

Unknowns and contradictions

  • The published evidence is a security advisory and patch, not a demonstrated production compromise. One-time authentication and explicitly unmanaged credentials are separate intended flows and should not be treated as identical to the managed-credential path.

Sources and citation

Material revision history

  1. Nov 15, 2026 · Canonical change recorded · new in release · revision 250