thethreatdeskFiled from the front lines of cybersecurity
BreachesHigh

Stolen Fireworks AI key led to a Cursor breach and a trojanized extension

Cursor says an access key exposed in a Fireworks AI breach let an attacker copy shared data and later push a backdoored Claude Code extension to up to 5,000 users.

Cursor, the AI coding editor, has explained how a recent attack on its systems began. An access key it had issued to AI inference provider Fireworks AI was exposed when Fireworks itself was breached.

What the attacker did

Using that key, the attacker copied data from a read-only storage bucket that Cursor used to share data with Fireworks. Credentials found in that data led the attacker into some of Cursor's software-as-a-service tools. Later, using a stolen credential, the attacker placed a trojanized build of Anthropic's Claude Code extension on the cache behind Cursor's extension marketplace on Oct 29. It was delivered as a normal update, tried to install a persistent SSH backdoor and stayed up for about 11 hours.

What was exposed

Cursor says Privacy Mode users, customer code repositories, its production environment and the Cursor app itself weren't affected. At most 5,000 users may have received the malicious extension.

The bigger picture

Uber and Notion have also disclosed data theft tied to the Fireworks breach, showing how one AI vendor can become a doorway into its customers.

Who's affected

Cursor users who installed Claude Code extension versions 2.1.312 or 2.1.313 on Oct 29, users who shared credentials with Cursor support, and other Fireworks AI customers.

What to do

If you installed the affected extension versions, treat the machine as compromised and rotate any credentials on it. Rotate API keys you've pasted into support chats or agent sessions, and review which AI vendors hold keys to your data.