A compromised access key linked to Fireworks AI enabled attackers to tamper with Cursor’s Claude Code extension, distributing a trojanized update that attempted to install a persistent SSH backdoor. The incident exposed a small fraction of user credentials and limited internal data, prompting Cursor to revoke the key and tighten its security posture.
How the breach unfolded
Cursor disclosed that an attacker obtained an access key originally issued to Fireworks AI. The key was unintentionally exposed after a separate breach at Fireworks, and the attacker used it to copy data from a read‑only storage bucket shared between the two services. That bucket held coding agent traces from sessions that were not protected by Cursor’s “Privacy Mode” setting. Most of the traces dated back to 2025 and were intended for model training.
Trojanized Claude Code extension
A trojanized version of Anthropic’s official Claude Code extension appeared on Cursor’s marketplace as a legitimate update. The malicious package attempted to install a persistent SSH backdoor on any system where the extension was installed. Users who accepted the update unknowingly opened a pathway for remote access.
Data that was accessed
The attacker leveraged the stolen credentials to reach several SaaS tools used by Cursor. The accessed data included:
- Select archived non‑production source code repositories, some containing business contact information
- Limited customer support data and email conversations
- Product usage metrics and contact details
- Summaries of certain coding agent conversations
- Credentials that users had sent to Cursor support
The exposure affected less than 0.01% of customers, according to Cursor, because only a tiny subset of sessions contained unredacted API credentials.
Cursor’s response
Cursor acted quickly after identifying the breach. The company:
- Deactivated the compromised vendor key
- Revoked and rotated all affected credentials
- Initiated a hunt for additional signs of unauthorized access
- Temporarily disabled internal and public extension publishing while the investigation continued
Why it matters to Indian readers
Developers and tech teams in India using AI‑assisted coding tools should review the security settings of any extensions they install. Enabling privacy‑focused options, such as Cursor’s Privacy Mode, can reduce the amount of data shared for model training and lower exposure risk.
What happens next / what readers should do
- Audit installed extensions: Verify that the Claude Code extension on your system is the official version released after the breach. Remove any suspicious updates.
- Enable privacy controls: Turn on Privacy Mode in Cursor to prevent session data from being used for training.
- Rotate credentials: If you used API keys or shared credentials with Cursor support, regenerate them and update any dependent services.
- Monitor for unusual activity: Watch for unexpected SSH connections or login attempts on machines that had the compromised extension.
- Stay informed: Follow Cursor’s official channels for further security advisories and apply patches promptly.
By taking these steps, Indian developers can mitigate the immediate risk and reinforce their broader security hygiene against supply‑chain attacks in the AI tooling ecosystem.
Sources
This report is based on information from:
