Cursor traces breach to Fireworks AI exposed key

Cursor has traced its October security breach to an access key exposed by Fireworks AI, revealing how a vendor-side leak led to a backdoored Claude Code extension reaching up to 5,000 users. The AI coding company published the findings in an incident update on 13 November, connecting the intrusion to the extension builds discovered on 29 October.
According to Cursor, an attacker used a Cursor-issued access key that Fireworks AI had exposed to copy data from a read-only storage bucket Cursor used to share data with the vendor. The bucket held limited coding-agent traces, mostly from 2025, from sessions not using Privacy Mode that Cursor was permitted to use for model training. A small number of those traces contained credentials that had not been removed, giving the attacker a route into selected software tools used by Cursor.
How far the exposure reached
Cursor said coding-agent traces were exposed for fewer than 0.05% of users who were not using Privacy Mode, and it identified and notified each affected user. Less than 0.01% of customers had API credentials exposed.
Using credentials found in the exposed records, the attacker reached selected software services used by Cursor. The material accessed included archived, non-production source repositories, limited support records and email conversations, product-usage metrics, contact details, summaries of some coding-agent conversations, and credentials that customers had sent to Cursor's support team. Current customer data and live traffic, active inference requests, and Privacy Mode prompts and code were not affected.
The backdoored extension
On 29 October, a credential stolen from the exposed internal data allowed the attacker to place modified Claude Code packages directly into the cache serving Cursor's VS Code extension marketplace. The affected packages were Anthropic.claude-code versions 2.1.312 and 2.1.313 for Windows and macOS. Cursor estimates that at most 5,000 users who requested specific cached versions potentially received a malicious build during roughly 11 hours of availability.
Cursor stressed that Anthropic did not publish the packages, that its product was not vulnerable, and that Open VSX, the marketplace source, never hosted the malware. Analysis of the malicious builds found injected JavaScript that ran when the extension loaded, without requiring a Claude conversation. It collected host details, added an SSH key labelled svc-telemetry, and attempted a reverse SSH tunnel to 134.209.82.225. It also set up persistence through a macOS LaunchAgent and a Windows scheduled task, retrying connections about every 60 seconds. Cursor said coding errors and missing administrator rights blocked remote login on most hosts, but did not prevent other malicious changes or host data collection.
What was not affected
Cursor reported no impact to the production systems and infrastructure that run its service, or to customer code repositories. Official app updates were unaffected, because release code sits in a separate hosting system the attacker never accessed, and no release-signing keys were exposed. The company validated all other extensions in its VS Code marketplace and reviewed and rebuilt that store.
Cursor's response
Cursor published indicators of compromise and remediation material on 30 October, and began notifying affected users and team administrators through in-app messages and email. Administrators received lists of specific keys to rotate where credentials were exposed, and Cursor asked the issuing third-party services to revoke them.
The company moved all inference traffic off Fireworks, deactivated the vendor key and cut off Fireworks' access to all datasets shared with it. Every exposed Cursor credential identified was rotated, followed by all credentials in use across internal systems as a precaution, even where there was no evidence of exposure. Cursor also expanded secret scanning, added network restrictions, redeployed production infrastructure and engaged external forensics experts. It believes the incident is contained: the last attacker activity it observed occurred on 1 November 2026 and was blocked, while monitoring continues.
Affected developers should follow Cursor's cleanup notice carefully. Removing the extension alone may leave SSH keys and startup scripts behind, so Cursor advises checking endpoint evidence, isolating affected machines, removing persistence, rotating accessible credentials from a trusted system, and assessing whether rebuilding is needed.
Fireworks' side of the story
Separately, Fireworks AI's own security bulletin reports that a third-party cloud provider hosting its research-and-development machines experienced an incident on 23 October 2026. An unauthorised party reached Fireworks development machines on 24 October and ran an unauthorised workload that led to additional customer-data impact. Fireworks detected the activity that day and says it has not detected further successful unauthorised access on the customer platform since 25 October.
Fireworks revoked and replaced credentials and service accounts, removed unnecessary elevated privileges, restricted network pathways and tightened storage and analytics access controls, with assistance from Palo Alto Networks Unit 42. Its investigation remains ongoing. The new account places the marketplace compromise at the end of a chain that began with a vendor key and continued through credentials left in historical traces.