Vercel Got Breached Through a Third-Party AI Tool — Here's Your 30-Minute Action Plan

April 20, 2026

This one broke yesterday. And it hits close to home for a lot of engineers.

Vercel confirmed a security incident involving unauthorized access to certain internal systems. The entry point wasn't a password leak or a brute-force attack. It was a compromised third-party AI tool called Context.ai — specifically, its Google Workspace OAuth app was the subject of a broader compromise. Once attackers had that foothold, they accessed select Vercel environments and read non-sensitive environment variables from a limited number of customer configurations.

A threat actor claiming to be ShinyHunters then posted on BreachForums, claiming access to internal databases, employee accounts, API keys, NPM tokens, GitHub tokens, and source code — asking $2 million for the lot. Vercel has confirmed the incident. The full scope is still being investigated.

The good news: Vercel says environment variables explicitly marked as Sensitive — which are encrypted and unreadable after being set — showed no evidence of access. The bad news: most people haven't marked their secrets as Sensitive.

There's an IOC you should check for immediately. If you or your team uses Google Workspace, audit for this OAuth app:

110671459871-30f1spbu0hptbs60cb4vsmv79i7bbvqj.apps.googleusercontent.com

Vercel is urging all Google Workspace admins to check for it now.

The Real Lesson Here

This is not just a "Vercel got hacked" story.

The breach came through a trusted integration. Not through Vercel's front door — through a side door that someone left open by connecting a third-party AI tool.

That's the uncomfortable part. An OAuth chain compromise means trust relationships become attack paths. When a tool you've connected to your workflow gets compromised, attackers may inherit access far beyond what you expected — GitHub repos, deployment pipelines, production environment variables, all of it.

I've seen a tweet making the rounds that captures it well: every SaaS tool your team connects is a security risk of its own — especially tools that request broad access to email, internal docs, and connected systems. Most modern AI tools ask for exactly these permissions. And most people approve the OAuth prompt in about three seconds.

That's the dangerous part. Not the hackers. The habits.

Every "Allow Access" click is a security decision. Your stack is only as secure as the most over-permissioned integration you forgot you approved.

Phase 1 — Lockdown (15 Minutes)

Don't start by rotating secrets. Start by stopping ongoing access.

Revoke old Vercel access tokens. Go to your account token settings. Keep only your active session and active CLI token. Revoke everything marked "Never Expires" — those are silent liabilities.

Scan your audit log. Check the last 7 days. Look for deploys you didn't trigger, environment variable reads, access from unfamiliar IPs, or permission changes you don't recognize. If something looks "probably fine," investigate it. Security mistakes hide behind assumptions.

Review installed integrations. This is the dangerous one. Go through every connected integration on your Vercel account. Remove anything you're not actively using — especially AI code review tools, vibe coding assistants, deploy automation, and third-party CI/CD integrations. Convenience creates attack surface.

Review GitHub OAuth apps. Most people use GitHub login for Vercel, which means GitHub is now part of incident response. Go to Authorized OAuth Apps and revoke aggressively. Six months unused? Remove it.

Rotate personal access tokens. Classic PATs with no expiration are delayed incidents. Revoke them. Create fine-grained tokens with minimal scopes and a 90-day expiration max.

Check your 2FA. If it's SMS-based, change it. SIM-swap attacks are still very real. Switch to an authenticator app or a hardware key. Also regenerate recovery codes — most people forget those exist until the worst possible moment.

Audit Google third-party access. Visit your Google account permissions. Remove old AI tools, browser extensions with account access, and random productivity utilities. Also check Gmail filters — attackers sometimes create silent forwarding rules so they keep receiving sensitive emails even after passwords are changed. That one gets missed often.

Phase 2 — Rotate Secrets (15 Minutes)

You don't need to rotate everything tonight. Start with your top one or two revenue-generating projects.

For each, rotate:

  • OpenAI, Anthropic, Groq API keys
  • Stripe secret key and webhook secret
  • Supabase service-role key and JWT secret
  • Database URLs containing passwords
  • Auth provider secrets — NextAuth, Clerk, Auth.js
  • Any third-party API keys connected to billing or customer data

Then go to Vercel → Project → Settings → Environment Variables. Edit every critical secret and mark it as Sensitive. Redeploy.

Sensitive variables are encrypted and can't be read back after being set. That alone significantly reduces blast radius if something goes sideways again.

Going Forward

Security isn't a one-time cleanup. It's maintenance.

A few habits that matter more than any single tool:

  • Mark all secrets as Sensitive in Vercel by default
  • Rotate important keys quarterly
  • Use minimum-scope tokens only — never "Full Access" when "Read" will do
  • Avoid "Never Expires" on anything
  • Audit third-party integrations monthly
  • Remove tools faster than you install them

The easiest security win is often deleting something.

The Bigger Picture

Most breaches don't start with elite hacking. They start with a forgotten integration, a PAT that was "just for testing," or an AI tool someone tried once and never removed.

This one started with a Google Workspace OAuth connection to a third-party AI tool that probably seemed useful at the time.

The supply chain angle is also worth watching. Vercel hosts Next.js, which sees around 6 million weekly downloads. If access to Vercel's internals were used to tamper with build pipelines, the blast radius extends well beyond Vercel's own customers. Vercel's CEO has confirmed Next.js and its open-source projects remain safe — but it's a good reminder that the platforms we trust to deploy code are themselves part of the software supply chain.

Take the 30 minutes tonight. It's probably the highest-ROI engineering task you'll do this week.

Source: Vercel April 2026 Security Bulletin · BleepingComputer

— Cheers, NP

GitHub
LinkedIn
Instagram