Case · DiggingBeagle record

ContextForge identity propagation could reveal identity data despite a disabled per-server setting

A ContextForge tool or resource request could carry gateway user identity to an upstream MCP server even when its specific identity-propagation setting was disabled.

Public disclosure: Nov 13, 2026

Scope

An authenticated agent gateway forwards tool/resource calls to an upstream server whose identity-propagation setting is off.

UnratedAI role: AGAINST AIAssessment method

At a glance

Mechanism and trust boundary

  1. 01

    Starting condition

    An authenticated agent gateway forwards tool/resource calls to an upstream server whose identity-propagation setting is off.

    Boundary: precondition and attacker capability

  2. 02

    Trust boundary crossed or evaluated

    The reported issue combines two boundaries: signed identity-claim material remained replayable without sufficient freshness and selected paths ignored the server-specific propagation opt-out. That exposes user identifiers, email addresses, group or team context to an unintended recipient under affected configurations.

    Boundary: mechanism

  3. 03

    Observed consequence

    A per-server privacy flag is not authoritative if the lower-level forwarding code still attaches replayable identity material.

    Boundary: observed effect and limitations

Claims & evidence

CLM-NOV15-CF110278-MECHA ContextForge tool or resource request could carry gateway user identity to an upstream MCP server even when its specific identity-propagation setting was disabled. The reported issue combines two boundaries: signed identity-claim material remained replayable without sufficient freshness and selected paths ignored the server-specific propagation opt-out. That exposes user identifiers, email addresses, group or team context to an unintended recipient under affected configurations.supported

Basis: reported finding

Link to claim
CLM-NOV15-CF110278-SCOPEThe advisory identifies CVE-2026-110278 and a 1.1.1 fix. A leak is conditional on the upstream path and configuration; no independently confirmed replay or third-party extraction was established by the cited report. This is not the same issue as credential override CVE-2026-110279. Claim replay and disclosure contrary to configuration require different tests from authorization to issue an upstream call.supported

Basis: reported finding

Link to claim

Implications

A per-server privacy flag is not authoritative if the lower-level forwarding code still attaches replayable identity material.

Controls and mitigations

  • Upgrade ContextForge and test identity propagation at each tool/resource route with a synthetic account.
  • Bind delegated identity claims to an intended upstream audience and short freshness window; inspect actual outbound headers, not only configuration state.

Unknowns and contradictions

  • This is not the same issue as credential override CVE-2026-110279. Claim replay and disclosure contrary to configuration require different tests from authorization to issue an upstream call.

Sources and citation

Material revision history

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