Cursor's November 13 incident report says an attacker used an access key exposed by its inference vendor Fireworks to copy coding agent traces, then reached support data, archived source code and credentials users sent to support. The same access planted the backdoored Claude Code extension on October 29. Fewer than 0.01% of customers had API credentials exposed; Privacy Mode users were not affected.
If you run a Cursor team, open the email Cursor sent your admins and rotate every key on its list. Cursor’s incident update, published November 13, says an attacker got in through an access key that Fireworks, its former training and inference vendor, exposed. Some of what was taken contained unredacted API credentials. If you never got a notice, Cursor says your data was not in the exposed traces.
What Cursor confirmed
The report fills in what was missing when the backdoored Claude Code extension surfaced on October 29. Cursor says the attack ran in this order:
- The vendor key. Cursor had issued Fireworks a key to a read-only storage bucket it used to share data. Fireworks exposed that key, and the attacker used it to copy the bucket.
- The bucket. It held coding agent traces, mostly from 2025, from sessions not in Privacy Mode. Fewer than 0.05% of users had traces exposed. Some traces held unredacted credentials, and fewer than 0.01% of customers had API credentials exposed.
- The SaaS tools. Credentials found in that data let the attacker into some of Cursor’s business tools. From there it took archived non-production source code repositories, limited support data and email conversations, product usage metrics and contact details, summaries of some agent conversations, and credentials users had sent to support.
- The extension. On October 29 the attacker used a stolen credential to place malicious copies of Anthropic’s Claude Code extension, versions 2.1.312 and 2.1.313, in the cache behind Cursor’s VS Code extension marketplace. They were up for about 11 hours, and at most 5,000 users could have received them. Cursor says Anthropic did not publish them and that this was not a flaw in Anthropic’s product.
Cursor says the last attacker activity it saw was on November 1 and was blocked, and it considers the incident contained.
What was not touched
Cursor lists these as unaffected: prompts and code from users in Privacy Mode, live traffic sent to Cursor today, production systems, customer code repositories, the Cursor app and its official updates (release signing keys were not exposed), other marketplace extensions and plugins, and its other products.
What to do
- Admins: work the rotation list. Cursor emailed team admins the specific keys to rotate and asked the issuing services to revoke them. Revocation by the issuer is not a reason to skip your own rotation. Check usage logs on those keys back to October.
- Anyone who pasted a secret into a support ticket: treat it as exposed and rotate it, whether or not you got a notice.
- Anyone who updated Claude Code in Cursor on October 29: run the check in Cursor’s extension notice. Cursor repeats that uninstalling alone may not remove the backdoor.
- Questions go to security@cursor.com.
Cursor says it has moved all inference off Fireworks, rotated every internal credential as a precaution, rebuilt the marketplace, added secret scanning to its repositories and brought in outside forensics experts.
Agent traces are now part of your attack surface
The detail that matters for every team using AI coding tools is where the credentials came from. Developers paste keys into agent sessions, agents read .env files, and those sessions get stored as traces for training and support. Cursor’s traces sat in a bucket shared with a vendor, and one leaked vendor key exposed them. Fireworks is the same vendor behind the Notion AI chat exposure disclosed on November 9. Two lessons for buyers. First, turn on Privacy Mode or the equivalent in every coding agent your team uses; in Cursor it was the line between exposed and safe. Second, keep secrets out of agent sessions entirely and use scoped, short-lived credentials, the approach behind 1Password for agents. Leaked AI keys get used fast: Railway customers saw $740,000 in charges and Northflank users saw stolen keys put to work within days. Cursor plan buyers can also read our note on Cursor’s X subscription.
