Vercel Breached by AI-Assisted OAuth Attack

🔑 Enhanced Key Takeaways

• •The breach originated from a sophisticated spear-phishing campaign targeting Vercel engineers, which bypassed traditional MFA by leveraging session token theft via an adversary-in-the-middle (AiTM) proxy.
• •Security researchers identified that the attackers utilized a custom-built LLM-based agent to automate the reconnaissance of Vercel's internal GitHub repositories, allowing them to identify and exploit misconfigured OAuth scopes in near real-time.
• •Vercel has initiated a mandatory rotation of all third-party integration tokens and is transitioning to a 'Zero Trust' architecture for internal service-to-service communication to mitigate the impact of future credential compromise.

🛠️ Technical Deep Dive

• •Attackers exploited 'over-privileged' OAuth scopes granted to internal CI/CD pipelines, specifically targeting the 'repo' scope which allowed unauthorized access to private source code.
• •The AI-assisted component involved a script that parsed leaked environment variables to identify valid API keys for third-party SaaS integrations, which were then used to pivot laterally within the infrastructure.
• •The breach utilized a 'Living off the Land' (LotL) technique, executing legitimate Vercel CLI commands to exfiltrate data, making detection by traditional signature-based EDR tools difficult.

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline