Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

Fireworks AI breach enabled trojanized Claude Code extension on Cursor Marketplace

Less than 0.01% of Cursor customers had API credentials exposed via unredacted agent traces.

A stolen access key exposed in a breach of Fireworks AI allowed an attacker to distribute a trojanized version of Anthropic's Claude Code extension through Cursor Marketplace, Anysphere disclosed in a Friday update1.

"An attacker obtained an access key we had issued to Fireworks," Cursor said. "The key was exposed by Fireworks, and the attacker used it to copy data from a read-only storage bucket we used to share data with them".

From data theft to supply-chain attack

The read-only storage bucket contained limited coding agent traces from sessions that were not in Privacy Mode, a Cursor-specific setting that lets users opt out of their data being used for training. Those sessions, most of which date back to 2025, were permitted to be used for model training. Among the traces were credentials that had not been edited out, exposing the API credentials of less than 0.01% of customers.

The attacker then used a stolen credential from the exposed internal-system data to place trojanized versions of Anthropic's Claude Code extension on the cache serving Cursor Marketplace on October 29, 2026. The malicious build was available for approximately 11 hours before Cursor removed it, and at most 5,000 users who requested specific cached versions of the extension on that date potentially received it. The trojanized extension attempted to install a persistent SSH backdoor on affected systems.

The attacker also used exposed credentials to reach select SaaS tools used by Cursor and access archived non-production Cursor source code repositories, some containing business contact information. Cursor said the attacker accessed limited customer support data and email conversations, product usage metrics and contact details, summaries of some coding agent conversations, and some credentials users had sent to Cursor support.

Containment and remediation

Cursor said it deactivated the vendor's key, revoked and rotated credentials, hunted for signs of deeper access, contained the incident after identifying further access, and temporarily disabled internal and product services. The company moved all inference traffic off Fireworks AI, deactivated the key issued to Fireworks AI, and cut off Fireworks AI's access to all data sets shared with it. Cursor notified affected users who installed versions 2.1.312 or 2.1.313 from the marketplace during the exposure window and helped revoke exposed credentials. No additional malicious activity had been detected since November 1, 2026, Cursor said.

There was no evidence the incident impacted users who had enabled Privacy Mode, customer code repositories, Cursor's production environment, the Cursor app and official updates, other apps, or any other extension or plugin downloaded from Cursor Marketplace.

Wider fallout from the Fireworks AI breach

The Fireworks AI compromise has rippled beyond Cursor. Uber confirmed that unknown threat actors stole thousands of drivers' license information after obtaining internal credentials; a service provider notified Uber on October 9, 2026, that an unauthorized third party had compromised the provider's internal credentials and used them to access Uber's account, and Uber's investigation determined on October 13, 2026, that the unauthorized party had accessed Uber data connected to the service provider. TechCrunch reported that the provider in Uber's breach was Fireworks AI. Notion also confirmed its data was stolen in the Fireworks AI hack.

ANALYSIS The incident illustrates how a single vendor-side key exposure can cascade into a supply-chain attack on a widely used developer tool, turning a data-access compromise into active code distribution through a trusted marketplace.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.