Request a Free Security Assessment
ZeroCeption Logo
ContactBlog
Back to Blog
Breach Analysis · April 21, 2026

The Vercel Breach: Six Posture Gaps That Turned One OAuth Click Into a $2M Data Leak

How a compromised AI tool, an unaudited OAuth grant, and a missing data exfiltration layer combined to expose customer secrets across one of the web's most trusted deployment platforms — and where Extended Security Posture Management closes every gap.

ZC
ZeroCeption Security Intelligence
XSPM Platform · zeroception.io
|
8 min read
TL;DR
  • Vercel was breached after an employee connected Context.ai to their enterprise Google account via OAuth. Context.ai was compromised via Lumma Stealer malware, and the attacker walked that OAuth trust path directly into Vercel's internal systems.
  • The root cause was not a single misconfiguration — it was six ungoverned posture gaps across identity, SaaS, AI tooling, cloud secrets, application dependencies, and data visibility.
  • The attack went undetected for 22 months. Credentials were in the wild 9 days before Vercel notified customers. The stolen data was listed for $2M on BreachForums.
  • Every gap in this breach is detectable and preventable through unified Extended Security Posture Management.
22 months
Dwell time undetected
9 days
Credentials leaked before disclosure
$2 million
Asking price, BreachForums

What Actually Happened

On April 19, 2026, Vercel — the deployment platform powering Next.js and thousands of production applications — confirmed unauthorized access to internal systems. The entry point was not a zero-day exploit in Vercel's infrastructure. It was a routine OAuth approval that one employee granted to a third-party AI tool called Context.ai.

The attack chain began months earlier and at a different company entirely:

February 2026

A Context.ai employee downloaded Roblox cheat scripts. Lumma Stealer malware harvested their Google Workspace credentials, API keys, Supabase tokens, and Datadog access.

March 2026

Attackers used the stolen Context.ai credentials to compromise their OAuth infrastructure, gaining access to tokens issued to Context.ai users — including a Vercel employee who had granted "Allow All" permissions.

March 27, 2026

The Context.ai Chrome extension was quietly removed from the Chrome Web Store. Simultaneously, dependency and supply chain signals tied to Context.ai infrastructure began appearing in threat feeds — unmonitored by anyone at Vercel.

April 19, 2026

The attacker used the hijacked OAuth token to access the Vercel employee's Google Workspace via SSO, bypassing MFA entirely. They enumerated and exfiltrated environment variables not marked as sensitive — API keys, database credentials, signing secrets — stored in plaintext internally.

April 19, 2026

A BreachForums post appeared under the ShinyHunters name, offering Vercel databases, employee accounts, access keys, and source code for $2 million.

"The techniques used are well-established. What makes this notable is that a single OAuth trust relationship cascaded into a platform-wide exposure for customers who had zero relationship with the compromised vendor."
— Trend Micro Research, April 2026

Root Cause: Six Posture Gaps

The Vercel breach was not caused by a single failure. It was the product of six distinct posture gaps — none requiring sophisticated exploitation, all preventable with continuous posture management:

Shadow AI in the stack An unsanctioned AI tool with enterprise-wide OAuth permissions. No inventory. No governance. No visibility.
GSuite OAuth misconfiguration "Allow All" permissions on a corporate identity provider, never reviewed. Sat open for 22 months without a single audit.
No identity anomaly detection The attacker moved laterally using a valid OAuth session for nearly two years. No behavioral signal was ever raised.
Plaintext secrets in the platform Environment variable encryption was opt-in, not default. A manual step nobody took became the primary exposure vector.
No vendor threat intelligence Context.ai was compromised in February 2026. Vercel disclosed in April. A 60-day warning window went entirely unmonitored.
No data exfiltration visibility No audit trail existed for what was read, copied, or sent out. Customers still do not know the precise scope of exposure.

Why This Breach Is Structurally Different

The industry commentary has largely focused on OAuth permissions hygiene — the employee should have granted scoped access, not "Allow All." That framing is not wrong, but it misses the structural dimension that makes this breach part of a larger pattern.

OAuth amplification. A single compromised trust relationship at a small AI vendor cascaded into a platform-wide exposure for Vercel customers who had no direct relationship with Context.ai. They simply used Vercel. The blast radius extended far beyond the initial OAuth grant.

AI-accelerated tradecraft. Vercel's CEO publicly attributed the attacker's unusual velocity to AI augmentation, describing their "surprising velocity and in-depth understanding of Vercel's systems." This is one of the first high-profile public attributions of AI-assisted adversary operations in a real-world breach.

The forensic blind spot. Google Workspace OAuth audit logs default to six months retention. The dwell time was 22 months. The forensic trail was cold before the investigation started.

Trend Micro places this within a 2026 convergence pattern alongside LiteLLM, Axios, Codecov, and CircleCI: attackers consistently go where developer credentials live, and use legitimate access paths to get there. The tooling layer — every OAuth-connected AI app, every SaaS integration — is the new perimeter. Most organizations have no continuous visibility into it.


Where ZeroCeption's XSPM Platform Changes the Outcome

Each of the six root-cause gaps maps directly to a detection and governance surface in ZeroCeption's Extended Security Posture Management platform. This is not a retroactive fit — it is precisely the threat class XSPM was designed to address.

AI-SPMShadow AI Discovery
The Gap

An unsanctioned AI tool with enterprise-wide OAuth permissions. No inventory. No governance. No visibility.

ZeroCeption Coverage

Continuous discovery and risk scoring of every AI tool connected to corporate identity. Unsanctioned integrations flagged, scoped, and routed to a governance review queue at the moment of authorization.

SSPMSaaS OAuth Governance
The Gap

"Allow All" permissions on a corporate identity provider, never reviewed. Sat open for 22 months without a single audit.

ZeroCeption Coverage

Inventory of every SaaS-to-SaaS OAuth grant, scored by permission scope and vendor posture. Over-permissioned grants surface in the governance queue with scope-reduction workflows and auto-revocation policy hooks.

ISPMIdentity Anomaly Detection
The Gap

The attacker moved laterally using a valid OAuth session for nearly two years. No behavioral signal was ever raised.

ZeroCeption Coverage

Behavioral baselines per human and non-human identity. New geos, new service-to-service calls, off-hours bulk reads — any drift against baseline triggers token freeze, forensic trace, and tenant-wide alerting.

CSPMSecrets & Cloud Posture
The Gap

Environment variable encryption was opt-in, not default. A manual step nobody took became the primary exposure vector.

ZeroCeption Coverage

Encryption-by-default enforcement across deployment secrets. Customer-scoped KMS keys, not platform-wide roles. Continuous drift detection against CIS, SOC 2, and internal baselines.

ZC-PulseVendor Threat Intelligence
The Gap

Context.ai was compromised in February 2026. Vercel disclosed in April. A 60-day warning window went entirely unmonitored.

ZeroCeption Coverage

Continuous dark-web and infostealer log monitoring against your vendor and employee domains. Credential IOCs match within hours and trigger automatic revocation of downstream OAuth grants tied to the compromised vendor.

DSPMData Exfiltration Visibility
The Gap

No audit trail existed for what was read, copied, or sent out. Customers still do not know the precise scope of exposure.

ZeroCeption Coverage

Access graph plus volumetric anomaly detection on every sensitive data store. Bulk reads of secrets, env vars, or customer records trigger immediate alerts and containment actions with full forensic replay.


Aurora AI

Aurora AI: Cross-Module Signal Correlation and Autonomous Remediation

Each module above surfaces its own findings. But the Vercel breach was not a single misconfiguration — it was a chain of signals that individually looked unremarkable and collectively described an active compromise. That is precisely the problem Aurora AI is built to solve.

Aurora AI is the intelligence layer across ZeroCeption's entire XSPM platform. It continuously ingests findings from AI-SPM, ISPM, CSPM, SSPM, ASPM, DSPM, and ZC-Pulse — and correlates them into a unified, contextualized threat picture that no single module can produce alone.

Cross-Module Signal Correlation

Aurora AI correlates signals across every module simultaneously. An over-permissioned OAuth grant (ISPM) + unreviewed SaaS integration (SSPM) + supply chain risk (ASPM) + active vendor IOC (ZC-Pulse) + plaintext secrets (CSPM) — individually medium severity, collectively a critical active threat narrative.

Enriched Threat Alerts

Rather than flooding teams with findings across six dashboards, Aurora AI produces a single enriched security alert: attack path, affected assets, implicated compliance frameworks, probable adversary objective, and a confidence score — the full threat picture in one view.

Dynamic Risk Rescoring

Aurora AI dynamically adjusts risk scores as new signals emerge. A medium-risk OAuth finding combined with an active vendor IOC from ZC-Pulse triggers automatic escalation to critical. Risk posture reflects the real threat in real-time — not the static output of the last scan cycle.

Autonomous Remediation

Where authorization has been granted, Aurora AI moves from analysis to action: revoking over-permissioned OAuth grants, triggering automated secrets rotation, enforcing encryption on misconfigured storage, and generating playbooks for findings requiring human approval — compressing response from days to minutes.

Applied to the Vercel Breach

In February 2026, ZC-Pulse ingests the Context.ai Lumma Stealer IOC from threat feeds. Aurora AI immediately correlates it against the active OAuth grant in ISPM and the unreviewed SaaS integration in SSPM — producing a single critical alert: "Vendor in your OAuth trust chain has an active IOC. Affected identity: [employee account]. Recommended action: revoke OAuth grant and rotate associated secrets." With autonomous remediation authorized, Aurora AI executes both in seconds. The attacker never gets the token. The breach never happens.

What To Do Right Now

If you use Vercel: Rotate all environment variables not explicitly marked sensitive immediately. Treat every non-sensitive env var as compromised until proven otherwise.

Check Google Workspace Admin → Security → API Controls for this OAuth app IOC:

110671459871-30f1spbu0hptbs60cb4vsmv79i7bbvqj.apps.googleusercontent.com

For every organization — regardless of Vercel usage:

  • Pull a full OAuth app audit in Google Workspace. Revoke every app you cannot justify with a current business reason.
  • Inventory every AI tool connected to your corporate identity via OAuth. Treat them as third-party vendors — because structurally, that is exactly what they are.
  • Ensure all secrets stored in deployment platforms are marked sensitive and encrypted at rest by default.
  • Enable 2FA and configure passkeys on all privileged accounts.

Conclusion

The Vercel breach is not a story about Vercel failing. It is a story about a threat class that most organizations have no continuous visibility into: the ungoverned layer of AI tools, SaaS integrations, and OAuth-connected third-party applications that employees wire into corporate identity systems every day.

None of the six gaps that enabled this breach required a sophisticated attacker. They required an attacker patient enough to wait for an opportunity that ungoverned posture creates automatically. The techniques were well-established. The window was left open by design assumptions, not technical failures.

Posture management is the discipline that closes those windows before they are used. ZeroCeption's XSPM platform — spanning AI-SPM, CSPM, SSPM, ISPM, ASPM, DSPM, and ZC-Pulse threat intelligence — provides the unified coverage model this attack class demands. And Aurora AI ties it together: correlating signals across every module, enriching alerts with full threat context, and — where authorized — remediating autonomously before the attacker has time to move. The question is not whether your environment has similar gaps. It is whether you have the visibility, and the intelligence layer, to find and close them first.

Sources
See what your OAuth graph looks like right now

ZeroCeption's XSPM platform gives you unified posture coverage across identity, cloud, applications, AI tooling, and SaaS — powered by Aurora AI.

Breach AnalysisSupply Chain SecurityOAuthShadow AIXSPMZero TrustISPMCSPMASPMAI-SPMDSPMPosture Management