Cursor Says Stolen Fireworks AI Key Led to Trojanized Claude Code Extension Incident

A Cursor access key exposed by Fireworks AI led to a data breach and a tampered Claude Code extension build that tried to plant an SSH backdoor.

cursor-fireworks-ai-key-trojanized-claude-code-extension

What’s new: Cursor has disclosed that an access key it had issued to inference provider Fireworks AI, which Fireworks then exposed, set off a chain of events ending in a trojanized build of the Claude Code extension on Cursor’s extension marketplace. The malicious build was a tampered copy of Anthropic’s official extension, served by Cursor’s marketplace as a legitimate update; the report doesn’t say Anthropic’s systems were breached. The trojanized build tried to install a persistent SSH backdoor.

How it unfolded

  • The attacker used the exposed key to copy data from a read-only storage bucket Cursor used to share data with Fireworks. It held coding-agent session traces from non-Privacy Mode sessions, mostly from 2025, some containing unredacted credentials (affecting under 0.01% of customers).
  • Those credentials gave access to Cursor’s SaaS tools, exposing archived non-production repositories, limited support data, usage metrics, some conversation summaries and credentials users had sent to support.
  • On October 29, 2026, the attacker used a stolen internal credential to place the trojanized build in the marketplace cache, where it stayed for about 11 hours. Cursor detected the last malicious activity on November 1.

Who’s affected

At most 5,000 users who received versions 2.1.312 or 2.1.313 of the extension from the marketplace on October 29. Cursor says Privacy Mode users, customer code repositories, its production environment, the Cursor app and official updates, and other marketplace extensions were not affected. The same Fireworks breach has been tied to data theft at Uber, which confirmed stolen driver license data, and at Notion.

What Cursor did

Cursor removed the build, took the marketplace offline, had the attacker’s server taken down, deactivated the Fireworks key, moved inference traffic off Fireworks, revoked and rotated credentials, and notified affected users.

What to do

  • If you installed versions 2.1.312 or 2.1.313 from Cursor’s marketplace around October 29, check for unexpected SSH keys, authorized_keys entries or persistence, and revoke any credentials that may have been exposed.
  • Avoid pasting secrets into AI coding sessions or support tickets, and turn on Privacy Mode where your tools offer it.
  • Treat keys issued to third-party AI providers as high-value secrets: scope them narrowly, monitor their use and rotate them regularly.

Sources