Cursor says its tampered-extension attack began with a key exposed at AI vendor Fireworks and credentials left in coding-agent traces
Cursor · Incident disclosure
Read the source (opens in a new tab)- What happened
- Cursor published an update on November 13, 2026 on the security incident behind the tampered Claude Code extension it disclosed October 29, after a two-week investigation.
- Key points
- Cursor says an attacker used an access key it had issued to Fireworks, "exposed by Fireworks," to copy coding agent traces, mostly from 2025, from a shared storage bucket.
- Some traces held credentials that had not been redacted; with them the attacker reached SaaS tools holding archived code and support data, then planted the extension.
- Under 0.01% of customers had API credentials exposed, and traces from sessions outside Privacy Mode were exposed for "fewer than 0.05% of users"; at most 5,000 users may have received the malicious build.
- Cursor moved all inference off Fireworks, rotated all internal credentials and added secret scanning; it says the last attacker activity it saw, on November 1, was blocked.
- Evidence
- The company's own incident account after an investigation supported by external forensics experts. It gives no total count of affected customers.
- Uncertain
- That Fireworks exposed the key is Cursor's account; the post includes no response from Fireworks. Cursor says it believes the incident is contained.