Back to feed

An update on our recent security incident

Cursor Blog

Nov 13, 2026

11/13/2026

Cache Delivery Compromise Allowed Malicious Claude Code Extensions To Be Served For About Eleven Hours

An update on our recent security incident · Cursor Blog

Science, Technology & Innovation · Nov 13, 2026

On October 29, an attacker used a stolen Cursor credential to modify the cache serving Cursor’s VS Code marketplace, distributing malicious copies of Anthropic’s Claude Code extension (versions 2.1.312 and 2.1.313) for about 11 hours. Up to 5,000 users may have been affected; this was not a compromise of Open VSX or Anthropic’s product, and affected users should follow Cursor’s system-check and remediation process because uninstalling alone may not remove the attacker’s changes.


11/13/2026

Cursor Conducts Comprehensive Containment And Remediation After Security Breach Rotates Credentials Rebuilds Infrastructure And Expands Monitoring To Reduce Persistence And Risk

An update on our recent security incident · Cursor Blog

Science, Technology & Innovation · Nov 13, 2026

Cursor contained unauthorized activity by disabling the vendor key and affected services, rotating exposed and broader internal credentials, moving inference traffic, rebuilding systems, expanding security controls, and using external forensics; these precautions caused temporary disruption but reduced persistence risk, while fewer than 0.05% of users had exposed non-Privacy-Mode traces and were notified.


11/13/2026

Vendor Shared Training Data Can Contain Secrets And Create Credential Risks In Production Adjacent Environments

An update on our recent security incident · Cursor Blog

Science, Technology & Innovation · Nov 13, 2026

A third-party Fireworks exposure let an attacker access historical Cursor coding-agent traces, where incompletely redacted credentials enabled limited access to selected SaaS tools and archived non-production data, highlighting the need to secure training datasets like production-adjacent systems.