Cursor’s account of the 29 October extension compromise is a relay race in which every baton was a credential.
On 13 November Cursor said an access key it had issued to Fireworks was "exposed by Fireworks", and an attacker used it to copy data from a read-only bucket the two firms shared. The bucket held coding agent traces, mostly from 2025, from sessions not in Privacy Mode, and some contained credentials nobody had stripped out: API credentials for fewer than 0.01% of customers, Cursor says. A transcript of an agent at work is, it seems, a credential store with better prose.
From there, Cursor says, the attacker reached select SaaS tools and, on 29 October, using "a stolen credential from the exposed data from our internal system", planted malicious copies of the Claude Code extension in the cache behind its marketplace. At most 5,000 users potentially received the build, which was up for about 11 hours.
That is Cursor’s account. Fireworks’ own security bulletins describe two incidents but do not mention Cursor, so I could not tie the key to either. Cursor says the last attacker activity it saw was on 1 November, and was blocked.
For buyers, "read-only" describes a permission, not what is inside, and agent transcripts now belong in the same inventory as the keys they leak. Cursor has since moved all inference traffic off Fireworks, one way to conclude a vendor review.
Sources: Cursor · The Hacker News · Fireworks AI