Malware targets 5,000 developers via manipulated Cursor extension
Attackers exploited leaked credentials from Fireworks AI to distribute a malicious version of the Claude-Code extension on the Cursor marketplace.
The short version
- Attackers used stolen Fireworks AI credentials to manipulate the Claude-Code extension distribution cache.
- Malicious versions 2.1.312 and 2.1.313 were available for a period of time on macOS and Windows.
- The malware included a JavaScript backdoor designed to establish a reverse-forwarding SSH connection.
- Less than 0.05 percent of users who were not in Privacy Mode had their programming assistant traces exposed.
Why it matters
This incident highlights how a single security failure at a third-party provider can ripple through the entire software supply chain. It serves as a reminder that developers must be cautious about the security of the tools and extensions they integrate into their workflows.
The Fireworks AI Breach
The security incident began with a breach at a cloud provider used by the AI company Fireworks AI. On October 24, 2026, Fireworks AI detected unauthorized access to its internal development machines. The attackers were identified as a group known for activities involving LLMJacking, a term for hijacking large language models.
Following the initial breach, the attackers obtained exposed access keys belonging to Cursor, a popular AI code editor. These keys were stored at the partner company, Fireworks AI, rather than being secured within Cursor's own primary infrastructure. This allowed the hackers to move laterally into the Cursor ecosystem.
By October 29, 2026, the attackers used these stolen keys to manipulate the distribution cache for the Claude-Code extension. They successfully copied a write-protected memory containing traces of programming assistants, most of which dated back to 2025. This maneuver allowed them to distribute compromised software to the broader developer community.
Malware Distribution and Mechanics
The attackers released manipulated versions of the Claude-Code extension, specifically versions 2.1.312 and 2.1.313, for both Windows and macOS. These malicious packages were hosted on the Cursor marketplace for a period of time, specifically between 2:27 and 13:35 PT on October 29. During this time, approximately 5,000 developers were at risk of downloading the infected files.
Analysis by Microsoft Threat Intelligence revealed that the manipulated JavaScript code contained a hidden SSH backdoor. This backdoor utilized a specific key labeled "svc-telemetry" to establish a reverse-forwarding connection. It was designed to transmit host details and attempt to establish a new connection every 60 seconds.
To ensure the malware remained active on a user's machine, the attackers used persistence mechanisms. They utilized LaunchAgents on macOS and scheduled tasks on Windows to keep the connection alive. However, the extension often failed to run successfully in practice due to various coding errors within the malicious package.
Impact on User Data
Cursor conducted a two-week investigation and concluded that the majority of its customers were not affected by the breach. The company noted that the impact was significantly contained due to the nature of the data accessed. Only a very small fraction of users who did not use the Privacy Mode were exposed to any risks.
For those who were affected, the attackers gained temporary access to internal SaaS tools, archived repositories outside of production, support data, and emails. They also viewed usage metrics, contact details, and summaries of programming conversations. These traces were primarily derived from unredacted access data that existed outside of the protected Privacy Mode.
The scale of the data compromise remained low according to the investigation results. Less than 0.05 percent of users who were not in Privacy Mode had their programming assistant traces exposed. Furthermore, the company reported that fewer than 0.01 percent of customers had their actual API access credentials compromised during the incident.
Remediation and Next Steps
Cursor took immediate action to neutralize the threat once the activity was detected. The company blocked the final actions of the attackers on November 1, 2026. They revoked the compromised Fireworks keys, rotated all exposed access credentials, and moved their inference processes away from Fireworks AI to improve security.
To further secure the platform, Cursor temporarily took its marketplace offline to rebuild it from the ground up. They also worked to shut down the servers used by the attackers and brought in external forensics experts to assist with the cleanup. The platform Open VSX was confirmed to have remained uncompromised throughout the event.
Security experts from GoPlus have recommended that any users who may have downloaded the affected versions use specific cleanup scripts. These scripts are designed to remove any potential remnants of the malware from the local system. Developers are encouraged to stay vigilant about the security of the extensions they install in their development environments.
What to watch
- How other AI code editors handle third-party extension security.
- Whether more developers will move toward 'Privacy Mode' defaults to avoid trace exposure.
Try it
Check if you have installed any Claude-Code extensions in the last two weeks and verify they are the latest official versions.
Sources and links
- ReportingBornCity: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by BornCity · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
