ExposureContext.ai and Vercel

An AI app’s OAuth grant became a way into Vercel

A Vercel employee had connected Context.ai, a third-party AI workspace, to their corporate Google account. When Context.ai was breached, the attacker used that access to reach Vercel’s internal systems and decrypt some customers’ environment variables.

By the NoFly team

What NoFly would decideIllustrative
Allowed, with a flagDiscover · OAuth grants
OAuth grant with broad Google Workspace access
Agent
Context.ai (Google Workspace app)
Came from
Approved by an employee
Rule
AI app with broad access to a work account
Why
The vendor reported a breach. Revoke the grant and rotate what it could reach.

What happened

Context.ai says it stopped unauthorized access to its AWS environment in March 2026, and that OAuth tokens for some users of its AI Office Suite were likely compromised. A Vercel employee had signed in to it with their enterprise Google account and granted it broad access.

The attacker used that access to take over the employee’s Google Workspace account and then their Vercel account, moved into a Vercel environment and decrypted customers’ environment variables that were not marked sensitive. Vercel disclosed the incident on 19 April 2026, told affected customers to rotate those values, and confirmed that no Vercel-published npm packages were tampered with.

  1. 01

    An employee connects an AI app to their work account with broad access.

  2. 02

    The AI vendor is breached and its OAuth tokens are taken.

  3. 03

    The attacker takes over the employee’s accounts.

  4. 04

    It reaches internal systems and decrypts customer environment variables.

Why it worked

Probable cause

A third-party AI app held broad OAuth access to a corporate account, so the AI vendor’s breach became a path into the company.

How NoFly stops it

  • Discover

    AI apps people granted access with their work account are found through your identity provider, with the scopes each one holds and who approved it.

  • Govern

    New AI apps stay marked new until someone reviews them, and a grant can be revoked in one step, or by an automation when the vendor appears in an advisory.

What to do this week

This week3 actions
  1. List AI apps connected to work accounts, and the scopes each one holds.
  2. Block broad scopes for unreviewed AI apps in your identity provider.
  3. Mark secrets as sensitive in your hosting platform so they cannot be read back.
Keep reading

More from the blog.

Find out what your AI is allowed to do today.

Book a call and we will map your agents, MCP servers and keys with you, or start on one laptop with the free scan.

Book a demo
npm i -g @nofly/agent