Cursor traced the intrusion to a leaked access key from Fireworks AI
Cursor、Fireworks AIによって露出したアクセスキーが侵害の原因と特定
In an incident update on November 13, 2026, Cursor traced the intrusion to a leaked Cursor-issued access key from Fireworks AI. The attacker used the key to copy historical coding agent run traces and credentials from a read-only storage bucket that Cursor shares with vendors, then used them to break into some of the SaaS tools it uses and distribute malicious Claude Code extension builds; fewer than 0.01% of customer API credentials were exposed; the bucket mainly contained run traces from 2025 sessions that had not enabled Privacy Mode, some containing unredacted credentials. On October 29 the attacker used the stolen credentials to place illicit builds into Cursor's VS Code extension marketplace cache; Cursor estimates that during a public window of about 11 hours, at most 5,000 users who requested the specific cached version may have received the malicious builds, and Anthropic did not publish these builds.
It lays out the attack chain from a vendor's leaked key to credentials in run traces and then to malicious extension distribution, which has lessons for vendor credential management.