Stolen AI Provider Key Used to Reach Cursor Data and Credentials
Cursor, the AI coding tool, has traced a security incident back to a third party. According to Dark Reading, a compromised key issued to AI infrastructure provider Fireworks was used by threat actors to access Cursor data and credentials.
The headline of the original report also links the incident to a malicious Claude Code extension. The short summary available to us does not explain how the extension relates to the stolen key, so we cannot say how the two connect or how far the access went.
What the summary does make clear is the route of the breach. The attackers did not need to break into Cursor directly. A key held in a supplier relationship gave them a way in, and that access exposed both data and credentials.
For small businesses that use AI coding assistants, extensions and hosted AI services, this is a reminder that those tools often depend on keys and tokens that sit with other companies. If one of those keys is stolen, the damage can reach your own accounts and data.
Why it matters
This incident shows that a key issued to a supplier can become the weak point in a product's defences, so a vendor's own security controls may not be enough to protect your data. Anyone using AI coding tools or extensions is exposed, particularly where credentials are stored or passed through those tools. It forces a decision about which AI services are trusted with credentials, and how fast those credentials can be rotated.
What to do
- Inventory the AI coding tools and extensions in use and remove any your team does not recognise or need.
- Rotate any credentials or API keys that were stored in, or passed through, Cursor.
- Check vendor security notices from Cursor for confirmed details on affected data.