Skip to content
Noroxi
OSV ecosystem advisory

GHSA-gqqj-85qm-8qhf: Paperclip: codex_local inherited ChatGPT/OpenAI-connected Gmail and was able to send real email

HighCVSS 8.7 · v3.1—No exploit No fix recorded

Noroxi assessment

Priority: Monitor (action score 34/100)

  • The sources we track (CISA KEV, Exploit-DB, Metasploit, Nuclei, GitHub) show no public exploit code or exploitation in the wild.
  • Anyone who can reach it over the internet can trigger it. A low-privileged account is enough. A user must open a link or visit a page. No special conditions are required; it is repeatable.
  • According to the record, no fixed version is available yet.

Low priority for now. Update when a fix is released; follow the vendor's advisory.

Generated automatically from the record's data: CVSS (NVD and CNA), CISA KEV and SSVC, FIRST EPSS, exploit sources, fixed versions. How we calculate it

Primary sources: OSV

### Summary A Paperclip-managed `codex_local` runtime was able to access and use a Gmail connector that I had connected in the ChatGPT/OpenAI apps UI, even though I had not explicitly connected Gmail inside Paperclip or separately inside Codex. In my environment this enabled mailbox access and a real outbound email to be sent from my Gmail account. After I manually intervened to stop the workflow, follow-up retraction messages were also sent, confirming repeated outward write/send capability. This appears to be a trust-boundary failure between Paperclip-managed Codex execution and inherited OpenAI app connectors, amplified by dangerous-by-default runtime settings. ### Details Successful runtime calls include: - `mcp__codex_apps__gmail_get_profile` - `mcp__codex_apps__gmail_search_emails` - `mcp__codex_apps__gmail_send_email` The connected Gmail profile resolved to my personal account. Inside the Paperclip-managed `codex-home`, I also found cached OpenAI curated connector state for Gmail under a path like: - `codex-home/plugins/cache/openai-curated/gmail/.../.app.json` This strongly suggests that the runtime had access to an already connected OpenAI apps surface rather than a Paperclip-specific Gmail integration that I intentionally configured. Separately, in the installed Paperclip code, `codex_local` defaults `dangerouslyBypassApprovalsAndSandbox` to `true`, and the server-side agent creation path applies that default when the flag is omitted. In practice, that makes this boundary failure much more dangerous because a newly created `codex_local` agent can operate with approvals and sandbox bypassed by default. The key issue is this: I had connected Gmail only in the ChatGPT/OpenAI apps UI. I had not intentionally connected Gmail inside Paperclip or separately inside Codex. Despite that, the Paperclip-managed `codex_local` runtime was able to use Gmail read/write actions. ### PoC Environment: - self-hosted Paperclip instance using `codex_local` - Gmail connected in the ChatGPT/OpenAI apps UI - no explicit Gmail connection configured inside Paperclip for this test - `codex_local` agent created and run with default behavior Observed reproduction path: 1. Connect Gmail in the ChatGPT/OpenAI apps UI. 2. Create or run a Paperclip `codex_local` agent. 3. Execute a task that inspects mailbox state or performs outward communication. 4. Observe successful Gmail connector calls such as: - `mcp__codex_apps__gmail_get_profile` - `mcp__codex_apps__gmail_search_emails` - `mcp__codex_apps__gmail_send_email` 5. Observe that the connected profile resolves to the ChatGPT/OpenAI-connected Gmail account and that mailbox reads and real sends are possible. Private evidence available on request: - successful `get_profile` / `search` / `send` logs - Paperclip-managed `codex-home` Gmail connector cache path(s) - screenshot showing Gmail write-capable actions such as `send_email`, `send_draft`, and `update_draft` exposed in the connected-app UI - incident timeline showing that a real outbound email was sent - recipient organizations, timestamps, message IDs, and sanitized evidence for both the original outbound email and the subsequent retraction messages ### Impact This was not only theoretical in my environment. It resulted in: - mailbox identity disclosure - mailbox search / thread access - a real outbound email being sent from a personal connected Gmail account to an external third party - follow-up retraction messages being sent after manual intervention, confirming repeated outward write/send capability From an operator/security perspective, connecting Gmail in the ChatGPT/OpenAI apps UI should not automatically make that connector available to a Paperclip-managed local agent runtime, especially not for write/send actions. One or more of the following: - no inherited OpenAI app connectors by default in Paperclip-managed `codex_local` runs - send/write connectors blocked by default - explicit Paperclip-side opt-in before outward actions - auditable approval and provenance for connector-mediated actions - safer defaults, including `dangerouslyBypassApprovalsAndSandbox = false`

Published
Apr 16, 2026
Updated
Apr 16, 2026
EPSS
—

Report tools

JSON

Action score

34

Monitor

Low priority for now.

CVSS
34 / 40 · 8.7 / 10
CISA KEV
0 / 30 · Not listed
EPSS
0 / 30 · —

Get told when this record changes

  • Not in KEV: no record of exploitation in the wild
  • No public exploit code on record
  • No fixed version yet

If any of these changes, we'll tell you in the daily email digest.

Follow GHSA-gqqj-85qm-8qhf

When you'll hear from us

  • when it enters CISA KEV
  • when exploit code appears or is weaponized
  • when the record is updated (including a fixed version)
  • when EPSS rises 20 points in a week

One daily email digest (06:00 UTC); connect a webhook for KEV listings and weaponized exploits every 10 minutes on paid plans, on the hour on the free plan.

Your free account also gets you

  • CVE alerts: add the products and versions you run, get the new CVEs that affect them
  • Uptime monitoring: 1 site, instant email when it goes down
  • Add your domain: subdomains and takeover risk scanned every week (1 domain)
  • 10 scans an hour on the security tools, at most 30 a day (3 a day without an account), an API key and MCP
  • All your scans in one panel: findings, what changed in the last scan, a report per domain or per client (PDF, share link)
All free tools
Is your version affected? Check it

Noroxi analysis

What does it mean in your environment?

The assessment above is generated automatically from the record's data; it doesn't know your environment. Our team verifies on your systems whether this vulnerability is actually exploitable and which mitigation is enough.

We use this product, ask for help

Affected systems

Is your version affected?

Pick the product and type the version you run. It's compared with this record's version ranges in your browser; the version isn't sent anywhere.

VendorProduct
npmpaperclipai

Package-level exposure

OSV and GitHub Advisory data: ecosystem, package and range. SBOM matching uses this table.

EcosystemPackageAffected rangeFix
npmpaperclipaiup to and including 2026.403.0—

Remediation

Which version to upgrade to

Fix versions compiled from the vendor, package registries and Microsoft. Verify the vendor's note before upgrading.

No fix version is recorded for this entry. Check the references for vendor advisories.

Exploit status

No known public exploit

No public exploit has been observed yet. That doesn't mean you're safe, only that the bar is a little higher.

Research context

For pentesters and researchers: attack profile, score disagreement, timeline, patch commits, credits, variant and chain candidates, bug bounty scope. All derived from existing data; no exploit code.

Timeline

From publication to today: proof of concept, Metasploit module, CISA KEV and fix record. Dates are as reported by the sources.

No dated events beyond publication.

Not on record for this CVE: EPSS history · patch commit or PR link · researcher credit · variant or chain candidate · public bounty program.

Technical details

Attack conditions

  • Anyone who can reach it over the internet can trigger it.
  • A low-privileged account is enough.
  • A user must open a link or visit a page.
  • No special conditions are required; it is repeatable.

If successful

Confidentiality
high · data can be read
Integrity
high · data or configuration can be modified
Availability
none

Impact can extend beyond the vulnerable component (scope changes).

Attack vector
Network
Attack complexity
Low
Privileges required
Low
User interaction
Required
Scope
Changed
Confidentiality impact
High
Integrity impact
High
Availability impact
None

CVSS vector

CVSS:3.1/AV:N/AC:L/PR:L/UI:R/S:C/C:H/I:H/A:N

osv

References

Show all references (1 more)

Vendor advisories and official records. Exploit/PoC links are deliberately left out.

All records