GitHub reviewed GHSA-rm8c-2cv8-jwf6 (CVE-2026-115770): Pocket ID before the July 27, 2026 backend build let stale email-verification tokens verify addresses the user never controlled, so OIDC responses could emit email_verified: true for a victim mailbox. If you run self-hosted Pocket ID for developer SSO — or link internal tools to it — upgrade the backend, audit recent account-linking events, and stop trusting email alone where sub is available.
Pocket ID is a lightweight identity broker some homelab and small-team deployments use instead of commercial IdPs. The bug is not OAuth crypto; it is state management around email verification tokens.
TL;DR
| Question | Answer |
|---|---|
| Am I affected? | Pocket ID backend before 0.0.0-20260727180059-43aaccd5bfac with default email verification settings |
| Is there a fix? | Upgrade to patched backend build; advisory includes regression test TestAuditEmailVerificationTokenSurvivesEmailChange |
| Attacker needs? | Normal user account, no admin, no victim mailbox access |
| Severity? | Critical per GitHub advisory |
OIDC apps using only sub? | Downstream takeover path does not apply per advisory |
| Agent relevance? | Auto-provisioned dev accounts tied to verified email are at risk |
Mechanism: token bound to user ID, not inbox
The advisory's summary, exercised through the HTTP API:
- Attacker signs in as
[email protected](an address they control). POST /api/users/me/send-email-verificationsends a token to that inbox.- Attacker does not redeem the token yet.
PUT /api/users/mechanges the account email to[email protected](unoccupied).- Pocket ID stores the new address with
emailVerified: falseinitially. POST /api/users/me/verify-emailredeems the old token that was emailed to[email protected].GET /api/users/menow shows[email protected]withemailVerified: true.- An OIDC flow with the email scope returns:
{
"email": "[email protected]",
"email_verified": true
}
Tokens are bound to user ID, not destination address, and address changes do not revoke outstanding tokens — the root defect.
Researchers verified against commit 9f559788a44cfea790de360da0a74d4bdf8ca5d5 and ship a Go regression test in the advisory (go test -tags=exclude_frontend,unit ./internal/service -run TestAuditEmailVerificationTokenSurvivesEmailChange).
Expected secure token lifecycle (what was missing)
Well-behaved identity providers usually bind verification artifacts to three tuples:
- User identifier (who requested verification)
- Destination address (where the link was sent)
- Issuance time (when the token was minted)
When any of email address, verification requirement, or account security state changes, outstanding tokens should invalidate. Pocket ID invalidated none of those on PUT /api/users/me, so step 6 in the reproduction used a token minted for an inbox the account no longer owned.
Downstream OIDC clients rarely see that bug directly — they only read email_verified: true — which is why the advisory frames impact as cross-application account takeover rather than Pocket ID session hijack alone.
Attack chain for application account linking
- Attacker creates Pocket ID user under their own email.
- Attacker requests verification, changes email to victim's corporate address (not yet registered).
- Attacker redeems stale token; Pocket ID marks victim email verified.
- Victim application trusts
email+email_verifiedto attach Pocket ID login to an existing local account. - Attacker completes OIDC login and inherits victim app account without mailbox access.
Apps that key solely on sub avoid this linking path — the advisory is explicit.
Who is affected
In scope:
- Teams running Pocket ID backend builds before the July 27, 2026 fix
- Relying parties that auto-provision developers from
email_verified - Homelab setups pairing Pocket ID with Gitea, Forgejo, internal dashboards, or agent UIs
Configuration defaults mentioned:
emailsVerified=false(verification on — default)allowOwnAccountEdit=true(users may change email — default)
Check and upgrade:
- Compare your deployed image or binary version to
0.0.0-20260727180059-43aaccd5bfac - Run the advisory's unit test against your checkout if you build from source
What this means if you run AI coding agents
Agents rarely talk to Pocket ID directly, but the SSO session agents inherit does:
- Git forge login determines which repos an agent may clone
- Internal model gateways often map OIDC groups to rate limits and keys
- CI OIDC federated roles may provision AWS access when email domains match
After patching Pocket ID, review audit logs for email changes followed quickly by verification. Rotate any API keys issued during suspicious sessions. Prefer sub-based account maps in new integrations.
Relate this to earlier Claude Code header email leaks we covered in Claude Code account email in request headers — different layer (transport vs IdP state), same lesson: email is a weak join key.
Patch verification and rollback planning
Upgrade path:
- Record current Pocket ID backend version from your container tag or
go versionbuild metadata. - Deploy
0.0.0-20260727180059-43aaccd5bfacor newer. - If you build from source, run the advisory's
TestAuditEmailVerificationTokenSurvivesEmailChange. On a vulnerable checkout it passes because it documents the bug; on a patched build the test should be updated or fail once tokens invalidate on email change. - Restart relying parties' OAuth caches if they cache JWKS or discovery documents aggressively (uncommon on homelab scales, but worth noting for sidecar proxies).
Post-patch audit:
- Search admin logs for
PUT /api/users/meimmediately followed byverify-emailfrom the same user ID with different email domains. - Review application logs where
email_verifiedflips from false to true without a corresponding mailbox challenge on the relying party. - For each suspicious linkage, rotate API keys and personal access tokens issued after that timestamp.
Policy hardening while you patch:
- Temporarily set
allowOwnAccountEdit=falseif your deployment supports it and you can tolerate admin-mediated email changes. - Disable auto-provisioning by email on satellite apps until Pocket ID is confirmed patched.
OIDC claim design for developer tooling
Modern coding products increasingly use OIDC for CLI device login and agent dashboard SSO. When you configure those integrations:
| Claim | Safer use |
|---|---|
sub | Primary stable identifier for account binding |
email | Display and audit only unless you re-verify out of band |
email_verified | Do not use alone for merge into existing local users |
Pair sub + issuer (iss) for uniqueness. If you must use email merges, require a second factor or admin approval when email_verified transitions on an account that already has repository access.
Agent orchestration tools that spawn per-user cloud roles should map sub, not email domain rules, to IAM trust policies — email domain rules are convenient for onboarding day one and painful on verification bugs.
Comparison to enterprise IdP incidents
Large commercial IdPs have shipped similar state-machine bugs around verification and recovery flows. Pocket ID's difference is deployment: self-hosted, often without a 24/7 SOC reviewing audit streams. Homelab operators may run it beside Forgejo, CI runners, and local LLM gateways agents call — concentrating risk on one small VM.
Critical severity in the advisory reflects OIDC trust amplification: one logical bug becomes many application account merges. CVSS details are on GHSA-rm8c-2cv8-jwf6; GitHub reviewed the entry on November 9, 2026 alongside several other backlog advisories.
Timeline
| Date | Event |
|---|---|
| 2026-07-29 | GHSA published to pocket-id/pocket-id |
| 2026-07-27 build | Patched pseudo-version 0.0.0-20260727180059-43aaccd5bfac |
| 2026-11-09 | GitHub Advisory Database reviewed/updated |
If your instance was internet-exposed before July 2026 without WAF or rate limits, assume opportunistic actors could have tried the API sequence even before public review — patch first, then audit.
Abuse scenarios beside account linking
The advisory's headline consequence is OIDC account takeover, but the same bad state can enable quieter harms:
- Support impersonation: Internal tools that display
email_verifiedin admin panels may grant attackers higher-touch support tiers or billing views tied to corporate domains. - Certificate or license issuance: Some self-hosted license servers email-gate feature flags; a verified victim domain can unlock entitlements without possession of the mailbox.
- Audit trail poisoning: Future investigators may trust Pocket ID audit exports showing the victim address as verified during attacker sessions, complicating incident timelines.
None of these require the victim to click a phishing link. They require only a relying party that treats IdP email claims as proof of employment.
Hardening checklist for homelab and small teams
Homelab Pocket ID deployments often sit beside single-node k3s, Ollama, and Claude Code or Codex workstations. Practical hardening without enterprise IdP budget:
- Patch to the July 27, 2026 backend build minimum.
- Expose admin APIs only on tailnet/VPN, not the public internet.
- Map every OIDC client and document whether it merges on
emailorsub. - Require manual admin approval for email domain changes on accounts that already have Git write access.
- Export weekly Pocket ID audit logs to storage agents cannot delete from the same laptop.
Coding agents should not store Pocket ID admin passwords in project .env files checked into git — see Claude Code don't-ask-again settings leaking to GitHub for how local agent configs become public.
Why GitHub reviewed the advisory in November 2026
The underlying flaw was published to the maintainer repository on July 29, 2026, but GitHub's Advisory Database reviewed the entry on November 9, 2026, alongside other backlog items. That delay does not change exploitability on unpatched instances exposed to the internet. Treat review date as a visibility signal for scanners, not as the start of exposure.
Security teams importing GitHub advisories into OSV or dependency scanners should ensure self-hosted Go modules that vendor Pocket ID are pinned to the patched pseudo-version, not merely "latest main." Pin the backend module hash in the same place you pin agent SDK versions so CI fails closed when the identity layer drifts.
What Beam can and cannot do here
Beam does not integrate with Pocket ID and does not validate OIDC token claims. If agents use OIDC refresh tokens stored in local keychains, Beam's redaction may limit how those tokens appear in hook logs, but it does not detect forged email_verified.
Where overlap exists: if an attacker uses a linked dev account to run Claude Code with stolen organization access, Beam's workspace policy can still flag risky tool calls on the workstation — secondary containment, not IdP repair.
Related reading
- Claude Code email in User-Agent headers
- Claude Code privacy mod still missing cases
- API keys in logs and environment variables
- Postman installer hooks targeting Claude Code and Codex
- GitHub advisory GHSA-rm8c-2cv8-jwf6
- NVD CVE-2026-115770
Frequently asked questions
What is CVE-2026-115770?
It is a critical flaw in Pocket ID backend before pseudo-version 0.0.0-20260727180059-43aaccd5bfac. Email verification tokens bind to user ID only, not the address mailed, and changing email does not revoke outstanding tokens, so attackers can mark arbitrary addresses verified.
What is the downstream impact?
OIDC claims can return email_verified true for an address the user never proved control of. Relying parties that link or provision accounts by verified email can merge the attacker Pocket ID subject into a victim application account.
What preconditions does the advisory require?
An ordinary Pocket ID account, email verification enabled (emailsVerified false, the default), allowOwnAccountEdit true (default), a target email not already assigned elsewhere, and a downstream app that trusts verified email for linking.
Which release fixes it?
The GitHub advisory lists patched pseudo-version 0.0.0-20260727180059-43aaccd5bfac and references verification against commit 9f559788a44cfea790de360da0a74d4bdf8ca5d5 before the fix.
Are apps that only use OIDC sub unaffected?
The advisory states applications that rely exclusively on the stable OIDC sub claim are not affected by the downstream account-takeover consequence, though you should still patch the identity provider.
Why does this matter for coding agent teams?
Self-hosted OIDC brokers often gate access to internal Git, artifact registries, and agent dashboards. Spoofed verified email can trick auto-provisioning flows that agents inherit through SSO.
API steps and version strings follow the GitHub advisory as of 2026-11-09.

