One Lumma Stealer download by a Context.ai employee became a kill chain that reached Vercel's internal environment through a compromised OAuth token — demonstrating that AI productivity tools with broad OAuth scopes create a supply-chain exposure most enterprises have not governed.
-
Context.ai employee infected with Lumma Stealer in February 2026 after downloading Roblox auto-farm scripts and game exploit executors
Verified 2026-05-21
-
Lumma Stealer is malware-as-a-service (MaaS) sold on underground forums since 2022
Verified 2026-05-21
-
Context.ai built a consumer product called AI Office Suite (launched June 2025), separate from enterprise product, deprecated after incident
Verified 2026-05-21
-
Attacker accessed Context.ai's AWS environment in March 2026 and exfiltrated OAuth tokens for AI Office Suite consumer users
Verified 2026-05-21
-
Vercel employee signed up for AI Office Suite using enterprise Google account, granted 'Allow All' permissions — this was shadow AI (not IT-sanctioned)
Verified 2026-05-21
-
OAuth tokens bypass MFA; they represent an already-authenticated session not requiring password re-entry
Verified 2026-05-21
-
Vercel: attacker pivoted from Google Workspace into internal systems via SSO, enumerated non-sensitive environment variables
Verified 2026-05-21
-
Rauch: attacker moved with 'surprising velocity' and 'in-depth understanding of Vercel's product API surface'; strongly suspects AI-acceleration
Verified 2026-05-21
-
Threat actor (using ShinyHunters name) posted on BreachForums selling stolen Vercel data for $2M; listing claimed 580 employee records, NPM tokens, GitHub tokens, partial source code
Verified 2026-05-21
How this was made. A.I. tools generated the animation, voice, and visual composition.
The facts, sourcing, and editorial decisions came from a human. Every claim in this video links to a
source above; every source was verified against its primary publication. The newsletter that started
this story is at
contextwindowsec.com.