Skip to content
Cybersecuritynews·Indexed by this site · Originally published· Event →· Its place in the scan record →SelectedAttention49

Cursor discloses how the leaked Fireworks AI key led to the Claude Code extension being backdoored

Cursor Reveals How a Leaked Fireworks AI Key Led to a Backdoored Claude Code Extension

AI summary

In a November 13 update, Cursor traced an October security incident to access keys leaked by Fireworks AI: the key Cursor issued to Fireworks could read-only access a shared storage bucket holding coding agent session records that had not enabled Privacy Mode and were approved for model training, some records had not had credentials scrubbed, and fewer than 0.05% of user records and fewer than 0.01% of customer API credentials were exposed. Using credentials in the records, attackers on October 29 placed a backdoored Claude Code package into Cursor's VS Code extension marketplace cache, involving Anthropic.claude-code 2.1.312 and 2.1.313 (Windows and macOS), and for about 11 hours up to 5,000 users requesting the specific cached versions may have received the malicious build.

Why it was selected

The leaked third-party key traveled from the storage bucket through internal credentials all the way to the extension marketplace cache; developers using Claude Code can check their machines for remnants against the affected versions.

Read originalcybersecuritynews.com