Cursor Breach and Fake Claude Code Extension: What to Check

On November 13, 2026, Cursor, the AI code editor, explained what caused its October security incident: a leaked key issued to a vendor exposed agent traces containing credentials, which led to a fake copy of the Claude Code extension in its marketplace. It matters to anyone who builds or maintains a website with AI coding agents.

What we know

  • How it started: according to Cursor, an attacker obtained an access key the company had issued to Fireworks, its inference provider, which Fireworks exposed. The attacker used it to copy data from a read-only storage bucket shared with that provider.
  • What was in it: coding agent traces, mostly from 2025, from sessions not using Privacy Mode. Cursor says some contained unredacted credentials and that fewer than 0.01% of its customers had API credentials exposed.
  • What happened next: with those credentials the attacker reached SaaS tools Cursor uses. From there, the company says, it accessed archived non-production repositories, support data, usage metrics, summaries of some agent conversations and credentials users had sent to support.
  • The fake extension: on October 29 the attacker placed malicious copies of the Anthropic.claude-code extension, versions 2.1.312 and 2.1.313, in the cache serving Cursor's extension marketplace. It was live for about 11 hours and at most 5,000 users could have received it, per Cursor. Anthropic did not publish that copy, and it was not a flaw in Anthropic's product.
  • What it did: Cursor's October 30 technical notice says it sent machine details to an attacker server and tried to set up persistent remote access on Windows and macOS.
  • Status: Cursor considers the incident contained. The last attacker activity it saw was on November 1. It moved off Fireworks, rotated credentials and rebuilt the marketplace.

What changes and what doesn't

According to Cursor, prompts and code from Privacy Mode users, customer code repositories, the Cursor app and its official updates, and every other marketplace extension were not affected. Among users without Privacy Mode, fewer than 0.05% had traces exposed, and Cursor says it contacted each one.

What changes is the lesson. The weak point was not the editor but a vendor and stored analytics data that held keys pasted into conversations. Fireworks published a bulletin on October 9 about its own incident involving internal credentials, detected on October 6; it does not mention Cursor, and Cursor's statement does not say whether it is the same event.

How to tell if this affects you

  1. Find out who uses Cursor on your site. Whether your developer is in-house or a contractor, ask if they use Cursor and whether they installed or updated the Claude Code extension on October 29.
  2. If they did, follow the technical notice. Cursor published checks and scripts for Windows and macOS. Uninstalling the extension may not be enough: treat that machine as compromised.
  3. Check the team admin's inbox. Cursor says it notified affected users in the app and by email, and sent team admins the list of keys they needed to rotate. If you got no notice, the company says your data was not in those traces.
  4. Rotate the keys that touch your site first. Server, database, payment gateway, AI API and CMS access, if any of them ever appeared in an agent chat or a support ticket.
  5. Turn on Privacy Mode. It made the difference here: sessions in that mode were not in the copied data.
  6. Never paste keys into the agent chat or support. Use environment variables or a secrets manager, and let the agent see the variable name, not the value.
  7. Pin extension versions. Turn off auto-updates on any machine with production access and check who publishes each extension before installing it.

Related: OpenCode Breach Put AI API Keys at Risk: What to Do

Sources

Updates: if Cursor or Fireworks publish more details, they will be added here with a link.