The Pulse
Cursor traces security incident to exposed Fireworks key
Cursor says an attacker used a key exposed by Fireworks to reach customer data and place a malicious Claude Code extension in its marketplace cache. At most 5,000 users potentially received the malicious build, according to the company.

An attacker used a key exposed by Fireworks to access data shared by Cursor and, later, place a malicious Claude Code extension in the AI coding tool’s marketplace cache, Cursor said. In a Nov. 13 update, the company said the incident exposed some customer data and potentially delivered the modified extension to at most 5,000 users.
The company estimates that less than 0.01% of customers had API credentials exposed. It said the vast majority of customers were unaffected and that it had contacted each affected customer directly.
How the attacker got in
According to Cursor’s investigation, the attacker obtained a key the company had issued to Fireworks and used it to copy data from a read-only storage bucket shared with the vendor. The bucket held limited coding-agent traces, mostly from 2025, from sessions outside Privacy Mode that Cursor was permitted to use for model training.
Some traces contained credentials that had not been redacted, which the attacker used to reach selected software services used by Cursor. From those tools, the attacker accessed selected archived, non-production Cursor source-code repositories, limited support data and email conversations, product-usage metrics, contact details, summaries of some coding-agent conversations and some credentials users had sent to support.
For users who were not using Privacy Mode, coding-agent traces were exposed for fewer than 0.05% of users, the company said. Cursor said the affected users had been identified and notified individually.
Malicious extension
On Oct. 29, the attacker used a stolen credential from the exposed internal data to place unauthorized copies of Anthropic’s Claude Code extension in the cache serving the Cursor VS Code extension marketplace, according to the update.
At most 5,000 users who requested particular cached versions that day potentially received the malicious build. The build remained available for approximately 11 hours before Cursor removed it, took the extension marketplace offline and worked with the attacker’s cloud provider to take the server offline.
Microsoft Threat Intelligence analyzed JavaScript injected into the package. The code attempted to add an attacker-controlled Secure Shell key, enable the local SSH service, establish reverse forwarding, configure persistence and report host details. Microsoft said successful remote access depended on host privileges, SSH configuration and the attacker’s server configuration; the code could still write keys and persistence artifacts even if tunnel access failed.
Cursor has asked customers who installed or updated to the tampered Anthropic.claude-code versions 2.1.312 or 2.1.313 through its VS Code extension marketplace during the exposure window to follow its security notice. The company cautioned that uninstalling the extension alone may not be enough.
Containment and response
The company believes the incident is contained and said the last attacker activity it had observed, on Nov. 1, was blocked. Its monitoring systems identified unauthorized activity, after which it deactivated the vendor key and began revoking and rotating credentials; some features were temporarily disrupted when it disabled internal and product services as a precaution.
Cursor said production systems, its app and official updates, customer code repositories, active inference requests, and prompts and code from Privacy Mode were not affected. The exposed traces were limited to storage used for historical offline analysis, it said.
The company moved all inference traffic off Fireworks, deactivated the key and cut off the vendor’s access to shared datasets. It also rotated exposed credentials and, as a precaution, credentials across its internal systems, rebuilt the VS Code extension marketplace, expanded scanning and network restrictions, and engaged external forensics experts.
Sources
Companies in this story
Related stories
- Cursor Finds Malicious Claude Code Extension in Its Marketplace
- Cursor rolls out shared Projects for teams in beta
- Replit Makes OpenAI Its Default, Keeps Anthropic’s Claude as an Option
- Lovable co-founder cites $600 million revenue pace at HumanX
- Claude Opus 5.5 (xhigh) Scores 18.5% on ProgramBench With mini-SWE-agent