Vercel, the cloud platform behind Next.js and widely used for enterprise web and frontend deployment, disclosed a breach yesterday after a threat actor gained access to internal environments and environment variables. The attack chain started several steps removed from Vercel itself. A Vercel employee had signed up for Context.ai, a small enterprise AI productivity tool, using their corporate Google Workspace account and granted it broad OAuth permissions. Context.ai was separately compromised in a breach traced to a Lumma infostealer infection at a Context.ai employee in February 2026. That initial infection gave attackers harvested OAuth tokens, which they used to take over the Vercel employee's Google Workspace account and pivot into Vercel's infrastructure. Environment variables not marked as sensitive were accessed, potentially exposing API keys, tokens, and deployment credentials. Vercel has confirmed the investigation is ongoing with Mandiant and other firms. The scope extends beyond Vercel: the same compromised OAuth app potentially affected hundreds of organizations whose employees granted it access to their Google Workspace accounts. Vercel published an indicator of compromise for the OAuth Client ID so administrators can check whether the app is present in their environment.

Cisco patched a critical vulnerability in Webex Services this week that allowed an unauthenticated remote attacker to impersonate any user in the service. The flaw, rated CVSS 9.8, existed in the SSO integration with Webex Control Hub. An attacker could supply a crafted token to a service endpoint and, due to improper certificate validation, gain access to legitimate Webex services as any user without providing credentials. Cisco addressed the vulnerability on the cloud infrastructure side, meaning the Webex platform itself has been updated. However, organizations using SSO still need to take one manual step: uploading a new identity provider SAML certificate through the Control Hub Alerts Center. Without that action, the SSO integration remains on the old certificate and the protection is incomplete. No exploitation in the wild has been confirmed by Cisco, but the combination of CVSS 9.8, unauthenticated access, and a required manual remediation step makes this worth elevating today. Cisco also patched three additional critical flaws in Identity Services Engine in the same advisory, with CVSS scores of 9.9, though those require authenticated admin access to exploit.

As of today, RedSun and UnDefend have no patches, no CVE identifiers, and no published remediation timeline from Microsoft. Both were introduced publicly by researcher Chaotic Eclipse on April 16, alongside BlueHammer, which Microsoft addressed in last week's Patch Tuesday as CVE-2026-33825. RedSun exploits how Windows Defender handles cloud-tagged files during its rollback mechanism. When Defender detects a cloud-tagged file and attempts to restore it, the attacker redirects that restoration to a privileged directory using NTFS junction manipulation. The result is SYSTEM-level code execution on Windows 10, Windows 11, and Server 2019 and later with near-complete reliability. UnDefend takes a different approach. Rather than escalating privileges directly, it silently blocks Defender from receiving signature updates while the endpoint continues to report as healthy to management consoles. An attacker using this combination runs UnDefend first to freeze Defender's threat intelligence, then follows with RedSun to reach SYSTEM. The sequence is designed to evade detection: from the SOC's perspective, the endpoint looks fully protected right up until it is not. Huntress has confirmed active exploitation of both techniques in observed incidents, with attackers staging the exploit files in user directories such as Pictures and Downloads.