Fireworks AI Breach Exposes Uber Driver Licences and Sparks Code-Extension Trojan
A Fireworks AI data breach leaked Uber driver licence details and enabled a trojanized Claude Code extension on Cursor’s marketplace, affecting thousands of users.
A security incident at AI-inference provider Fireworks AI has led to the theft of Uber driver licence data and the distribution of a trojan-laden version of Anthropic’s Claude Code extension on Cursor’s marketplace. The breach began when Fireworks exposed internal credentials that were later used to access Uber’s data, and the same compromised key was leveraged by attackers to infiltrate Cursor’s infrastructure.
- Uber drivers’ licence information was stolen after Fireworks AI’s credentials were compromised.
- Cursor’s marketplace hosted a malicious build of the Claude Code extension for up to 11 hours, potentially affecting 5,000 users.
- Both Uber and Cursor have taken immediate remedial actions, including key revocation, credential rotation, and disabling affected services.
How the Breach Unfolded
On 9 October, Fireworks AI notified Uber that an unauthorized party had stolen internal credentials. Those credentials were used to reach Uber’s account with Fireworks, allowing the attackers to pull personal data of drivers. Uber filed notices with state attorneys general, listing 2,567 Massachusetts, 3,430 South Carolina, and 1,873 Rhode Island residents affected – a total of 7,870. California and Illinois were also notified, but no numbers were released, so the figure is likely higher.
In a separate incident, the same compromised key was used by attackers to copy data from a read-only storage bucket that Cursor shared with Fireworks. The bucket contained coding-agent traces from sessions that were not in Privacy Mode, including unredacted API credentials. Cursor reported that less than 0.01% of its customers had credentials exposed in this way.
Cursor’s Trojan-laden Extension
On 29 October, attackers used the stolen credential to place trojanized versions of Anthropic’s Claude Code extension on the cache that serves Cursor’s extension marketplace. Users who requested specific cached versions of the extension between 29 October and 1 November could have received the malicious build. Cursor estimates that at most 5,000 users were affected during the 11-hour window. The trojan attempted to install a persistent SSH backdoor on affected systems.
Cursor immediately deactivated the Fireworks key, revoked and rotated credentials, and disabled internal and product services to contain the breach. The company also notified users who installed versions 2.1.312 or 2.1.313 and helped them revoke exposed credentials. No further malicious activity has been detected since 1 November.
Impact on Uber Drivers and Other Customers
Uber confirmed that thousands of driver licence records were stolen, but the company has not disclosed the exact number of affected individuals beyond the state notices. The data was obtained through Fireworks’ internal credentials, which were also used to access billing data and, for some customers, secrets, datasets, and unfinished model weights. Notion, another client of Fireworks, also confirmed that AI chat transcripts from fewer than 1% of its workspaces were exposed.
Fireworks, valued at $17.5 billion as of July, has identified two incidents in October: one involving credential misuse and another involving a breach at a cloud provider that allowed attackers to reach its development machines. The company attributes the attacks to a group that sells stolen AI credentials, a practice known as LLMjacking.
What This Means for Developers and Businesses
The incident underscores the risks of third-party AI vendors handling sensitive data. Developers relying on AI inference services should verify that vendors have robust credential management and monitoring. Businesses that integrate AI models must enforce strict access controls, rotate keys regularly, and audit third-party access logs. The fact that a stolen key enabled a trojanized extension also highlights the importance of securing extension marketplaces and monitoring for anomalous code deployments.
Both Uber and Cursor have taken decisive steps: Uber notified state authorities and is investigating the scope of the breach, while Cursor has revoked the compromised key, rotated credentials, and temporarily disabled affected services. The companies have also worked with cloud providers to remove malicious builds and ensure no further compromise.
As AI services become more integral to software development, this breach serves as a cautionary tale. Organizations must balance the convenience of AI tools with stringent security practices to protect user data and maintain trust.



