← Back to blog

Stolen AI Key Led to Compromised Claude Extension

A stolen access key from Fireworks AI enabled attackers to distribute a trojanized version of Anthropic's Claude Code extension on Cursor's marketplace. This incident highlights risks in third-party AI integration and credential exposure.

TL;DR

  • Attackers used a compromised Fireworks AI access key to tamper with the Claude Code extension.
  • The breach originated from credentials exposed during a prior Fireworks AI security incident.
  • Cursor confirmed the extension was replaced with a malicious version on their marketplace.
  • Organizations using AI-powered development tools should audit third-party integrations.
  • Proper credential rotation and access control could have prevented this supply chain attack.

In a recent supply chain security incident, threat actors leveraged a compromised access key from Fireworks AI to distribute a malicious version of Anthropic's Claude Code extension through Cursor's extension marketplace. The breach underscores the cascading impact of credential exposure and the growing attack surface introduced by third-party AI integrations in developer tools.

Cursor disclosed that an attacker gained unauthorized access by exploiting an access key previously issued to Fireworks AI. According to Cursor, the key was inadvertently exposed following a security breach at Fireworks AI, which the attacker then used to manipulate data and ultimately replace the legitimate extension with a trojanized variant.

Attack Vector and Timeline

  • The initial compromise occurred when Fireworks AI suffered a breach that led to the exposure of access keys.
  • One of these keys, originally granted to Cursor for integration purposes, was accessed by an attacker.
  • Using the stolen key, the attacker copied data from a read-only environment and later manipulated the Claude Code extension.
  • The trojanized extension was uploaded to Cursor's extension marketplace, potentially affecting users who downloaded it.

Security Implications for Developer Platforms

  • This incident demonstrates how breaches at third-party services can directly impact downstream platforms like Cursor.
  • Exposed API keys and credentials remain a critical vulnerability, especially when they grant access to trusted integrations.
  • Supply chain attacks targeting AI-enhanced developer tools pose significant risk due to deep IDE integration and code execution capabilities.
  • Organizations must enforce strict credential lifecycle management and monitor third-party access permissions regularly.
  • Prompt detection and revocation of compromised keys could have limited the scope of this attack.

Sources

Sources

Security email updates

One digest email when we publish new security articles (TL;DR plus links to read more). Unsubscribe anytime from the message footer. See our Privacy Policy.

Stolen AI Key Led to Compromised Claude Extension — Agent Breach Blog | Agent Breach