Cursor says an October security incident exposed some older AI coding sessions, API credentials and customer support records. In a Nov. 13 report, the company said it had contacted affected customers directly. It began sending in-app and email notices on Oct. 30, and emailed team administrators a list of specific credentials to replace where it had identified exposed keys.
The report adds a second concern to the one Cursor disclosed in October. That warning centered on a tampered Claude Code extension offered through Cursor’s VS Code extension marketplace. The new findings cover records and credentials that could affect a team even if nobody installed the extension.
What the investigation found

Cursor said someone copied historical records from storage it had shared with Fireworks AI, a service provider. The records, mostly from 2025, came from sessions in which users had turned off Privacy Mode and Cursor was permitted to use the material for model training. These records, which Cursor calls traces, capture parts of an AI coding session. Some contained API credentials that had escaped redaction.
Cursor reported that fewer than 0.05% of users had coding-agent traces exposed, and fewer than 0.01% of customers had API credentials exposed. Those figures describe two specific kinds of exposure. The company did not give a single count for all impacted customers. It also reported access to limited support data and emails, some coding conversation summaries, business contact details, usage metrics and some credentials that users had sent to support.
In its separate incident bulletin, Fireworks reported access to certain customers’ dataset and model-weight material and said it was contacting affected customers. Cursor said it revoked the key it had issued to Fireworks and moved live inference traffic off the provider. Cursor said the copied files came from storage for historical analysis. The company reported no exposure of live coding requests, customer code repositories or official Cursor app releases, and said it blocked the last unauthorized activity it observed on Nov. 1.
What a notice calls for

Start with the category in the notice. A support-data notice can cover a conversation or contact details. A credential notice calls for action with the service that issued the key. If you use Cursor through work, pass the notice to the team administrator or security contact. Administrators can check the list Cursor sent against the systems their team owns. Open Cursor directly to look for an in-app notice if you are unsure about a message in your inbox.
An API key lets an application authenticate to a service. If Cursor identifies one as exposed, revoke it with the issuing service and replace it wherever your organization still needs access. A new Cursor account password does not disable a key issued by another provider. GitHub’s guidance on exposed credentials explains why deleting the text that contains a key leaves the credential active until it is revoked. If you received an extension notice, follow Cursor’s incident instructions for that affected installation. Cursor says removing the extension alone may leave work to do.
What Privacy Mode covers
Cursor said code and prompts from users working in Privacy Mode were outside the exposure it described. That finding applies to those sessions. A team can still receive a separate notice about data it shared in a support thread or about an extension installation.
Cursor’s current data-use rules explain that, with Privacy Mode off, the company can store and use prompts and code data to improve its products and train models. Teams can enforce Privacy Mode for members, and Cursor documents separate admin controls for the few models that require provider retention. Review the team setting where people actually work, especially if employees can switch between work and personal accounts.
.jpg)





