AI-assisted development platform Cursor has confirmed that a supply chain incident resulting in the distribution of a malicious extension was triggered by an access key exposed during a breach at third-party vendor Fireworks AI. The compromise permitted attackers to poison Cursor's extension cache and distribute a backdoored version of Anthropic's official Claude Code extension.
According to an investigation summary reported by The Hacker News, the threat actor leveraged the exposed vendor key to exfiltrate training traces from a shared storage bucket, eventually pivoting into Cursor's internal systems and software-as-a-service infrastructure.
From Exposed Bucket to Internal Pivot
The incident began when attackers acquired an access key that Cursor had previously issued to Fireworks AI. Using this key, the adversary exfiltrated information from a read-only storage bucket used to share data between the two companies.
Cursor stated that the bucket housed limited coding agent session traces, primarily from 2025, from users who had not activated Cursor's "Privacy Mode." While this data was permitted for use in model training, unredacted credentials were inadvertently present in the records, exposing API credentials belonging to less than 0.01% of Cursor's user base.
The attacker subsequently used these exposed credentials to compromise several internal SaaS tools utilized by Cursor. Through this foothold, the actor accessed:
- Archived non-production Cursor source code repositories containing some business contact details.
- Summaries of certain coding agent conversations.
- Limited customer support records, email threads, and credentials submitted to the support team by users.
- Product usage metrics and user contact information.
Cursor responded by deactivating the vendor key, rotating and revoking exposed credentials, and temporarily disabling select internal services and product features to halt lateral movement.
Cache Poisoning Distributes SSH Backdoor
Despite containment efforts, the attacker leveraged a credential retrieved from the exposed internal systems to stage a secondary attack on October 29, 2026. The threat actor accessed the cache responsible for serving the Cursor Marketplace and injected trojanized builds of Anthropic's official Claude Code extension.
The malicious builds—identified as versions 2.1.312 and 2.1.313—were designed to install a persistent SSH backdoor on victim machines. Cursor estimated that up to 5,000 developers who requested the extension during an 11-hour exposure window received the malicious package.
Once detected, Cursor removed the poisoned builds, took the extension marketplace offline, and coordinated with the cloud infrastructure provider hosting the attacker's server to shut it down. The company confirmed that no malicious activity has been observed on its infrastructure since November 1, 2026.
Vendor Breach Impacts Broader Tech Ecosystem
The blast radius of the Fireworks AI security failure extends beyond Cursor. Uber submitted a breach disclosure with authorities in California noting that on October 9, 2026, a service provider informed the company of compromised internal credentials, which were later used to steal the driver's license details of thousands of Uber drivers.
While Uber's regulatory filing did not name the third party, industry reports identified Fireworks AI as the compromised provider. Productivity software startup Notion has also confirmed that its data was caught up in the same breach.
Cursor noted that customer code repositories, production environments, official application updates, and other marketplace plugins were not impacted by the incident. Users who operated with Privacy Mode enabled were also unaffected by the bucket leak.
What You Should Do
Development and security teams using Cursor should implement the following defensive actions immediately:
- Inspect Installed Extensions: Check local development environments for the Claude Code extension. Verify whether versions 2.1.312 or 2.1.313 were installed, specifically on or around October 29, 2026.
- Audit SSH Configurations: Examine local and user-level SSH files (including
~/.ssh/authorized_keys) on machines where Cursor was used during the exposure window to identify any rogue public keys or unexpected backdoor access paths. - Rotate Sensitive Secrets: Immediately revoke and replace any local environment variables, SSH keys, private tokens, and API credentials stored or accessed on systems that received the compromised extension.
- Enforce Privacy Controls: For organizations utilizing Cursor, ensure that Privacy Mode is enabled company-wide to prevent developer session traces and inadvertently copied credentials from entering vendor training datasets.





