
Cursor has disclosed that the distribution of a compromised version of Anthropic’s official Claude Code extension through its marketplace was connected to the theft of an access key exposed during a security breach involving Fireworks AI.
The AI-powered coding platform said an attacker obtained an access key it had issued to Fireworks AI. The key was subsequently exposed following a breach at the AI infrastructure provider, allowing the attacker to extract information from a read-only storage bucket Cursor used to share data with the company.
According to Cursor, the stolen key enabled the attacker to copy data from the storage bucket, which contained limited records of coding-agent sessions that were not protected by Privacy Mode.
The disclosure follows an incident in which a trojanized version of Anthropic’s official Claude Code extension appeared on Cursor Marketplace as a legitimate update. The malicious extension attempted to install a persistent SSH backdoor on affected computers.
Cursor explained that the exposed storage bucket contained coding-agent traces, mostly dating from 2025, which had been approved for model-training purposes. Some of the records contained credentials that had not been properly removed, exposing API credentials belonging to fewer than 0.01 per cent of Cursor customers.
Privacy Mode is a setting that allows Cursor users to prevent their data from being used for model training.
The attackers reportedly used the exposed credentials as an entry point to access selected software-as-a-service platforms used by Cursor, gaining access to several categories of company and customer information.
The compromised information included selected archived repositories containing non-production Cursor source code, some of which held business contact details. Other exposed records included limited customer support information, email conversations, product usage statistics, contact details, summaries of certain coding-agent interactions and credentials submitted by users to Cursor’s support team.
Cursor said it responded by disabling the compromised vendor access key, revoking and rotating credentials, and investigating whether the attackers had gained access to additional systems.
The company also temporarily shut down some internal and product services while working to contain the incident, resulting in temporary disruptions to certain Cursor features.
The attack subsequently escalated on October 29, 2026, when the intruder used credentials obtained from the exposed internal data to upload malicious versions of Anthropic’s Claude Code extension directly to the cache serving Cursor’s extension marketplace.
Cursor estimated that up to 5,000 users who requested particular cached versions of the extension on October 29 may have received the compromised software.
The malicious versions remained available for approximately 11 hours before the company discovered and removed them. Cursor subsequently took its extension marketplace offline and worked with the cloud hosting provider responsible for the attacker’s server to have the infrastructure taken down.
Users who installed versions 2.1.312 or 2.1.313 of the extension during the affected period were notified. Cursor also assisted affected customers in revoking credentials that may have been exposed.
As part of its security response, the company moved all inference traffic away from Fireworks AI, deactivated the access key issued to the provider and terminated its access to datasets previously shared with it.
Cursor said it had not detected further malicious activity since November 1, 2026.
The company also stressed that there was no evidence the incident affected users who had enabled Privacy Mode, customer code repositories, its production environment, the Cursor application or its official updates. Other Cursor applications, extensions and plugins downloaded from its marketplace were also not believed to have been affected.
Uber Also Affected by Fireworks AI Security Breach
The disclosure comes amid reports that Uber also suffered a data breach linked to compromised credentials belonging to an external service provider.
Uber confirmed that unauthorised individuals accessed information connected to its account with the provider, including thousands of drivers’ licence records.
In a filing with authorities in California, the company said the service provider notified it on October 9, 2026, that an unauthorised third party had compromised its internal credentials and used them to access Uber’s account.
Uber’s investigation, completed on October 13, established that the attackers had used the compromised credentials to access data associated with the company’s relationship with the provider.
Although Uber did not identify the service provider in its disclosure, TechCrunch reported that it was Fireworks AI.
The company has not revealed the exact number of affected users.
The Fireworks AI incident has also been linked to a separate data theft involving online productivity startup Notion, which confirmed last week that some of its information had been stolen during the breach.
The incidents highlight the security risks organisations face when third-party service providers have access to internal systems, customer information and sensitive credentials. A compromise affecting one supplier can potentially expose several businesses that depend on its services.



