Cursor, the maker of an AI code editor, says attackers reached its systems through Fireworks AI, a company that runs AI models for it. According to BleepingComputer, the attackers used an access key stored at Fireworks to copy data that held unredacted credentials. On October 29, they used one to place tampered copies of Anthropic's Claude Code extension, an add-on for the AI coding assistant, on Cursor's extension marketplace cache. The copies stayed up for about 11 hours, and up to 5,000 developers may have received them. Researchers say coding errors likely kept the malicious code from working on many machines.
The case shows how one AI provider's breach can spread into its customers' software supply chain. Cursor says its production systems, official releases and customer code repositories were not affected. The company says it rotated credentials, moved all inference traffic off Fireworks and rebuilt the marketplace.
