RC RANDOM CHAOS

Fireworks Breach Cascaded to Malicious Cursor Extensions

· via Dark Reading

Original source

Cursor Traces Malicious Claude Code Extension to Third-Party Breach

Dark Reading →

A malicious version of Anthropic’s Claude Code extension appeared in Cursor’s marketplace for approximately 11 hours on October 29, potentially reaching around 5,000 users. The tampered builds attempted to harvest system information and establish persistent remote access via attacker-controlled SSH keys. Cursor quickly removed the malicious versions, shut down its marketplace temporarily, and conducted a comprehensive audit.

Investigation traced the attack to a compromised access key issued to Fireworks, an AI infrastructure provider. Fireworks itself had been breached on October 23 when attackers targeted its R&D systems. Using Fireworks’ stolen credentials, attackers accessed Cursor’s read-only storage containing historical coding-agent session records—some with improperly redacted credentials. The attackers then leveraged those credentials to infiltrate Cursor’s extension cache system and inject the malicious builds.

The incident exemplifies escalating supply chain risks in interconnected development tool ecosystems. Attackers apparently targeted Fireworks as part of LLMjacking operations—monetizing unauthorized access to large language models. Though exposure was limited (fewer than 0.01% of customer API credentials, 0.05% of session records), the breach demonstrates how compromise at a single vendor can cascade downstream across multiple systems. Similar attacks have recently surfaced, including hundreds of exposed secrets in VS Code extension repositories and malicious GlassWorm extensions deployed to steal developer credentials.

Read the full article

Continue reading at Dark Reading →

This is an AI-generated summary. Read the original for the full story.