Cursor Says Fireworks Compromise Led to Malicious Claude Code Extensions

Summary
Cursor says attackers used credentials exposed in the Fireworks AI compromises to access Cursor systems and serve tampered Claude Code extensions through its marketplace for about 11 hours, potentially affecting up to 5,000 users.
Key points
- Cursor says attackers used a compromised access key issued to Fireworks to access a read-only storage bucket and obtain unredacted credentials for Cursor SaaS tools.
- The attackers accessed some archived non-production source code, limited support data and email conversations, product usage metrics, and other information, according to Cursor.
- Two tampered Claude Code extension versions, 2.1.312 and 2.1.313, were served from Cursor's marketplace cache for roughly 11 hours.
- Cursor estimates at most 5,000 users may have been affected; the malicious versions were not present in the Open VSX marketplace source.
- Cursor says it blocked the access after discovering the activity and has seen no attacker activity in its environment since November 1.
Article Details
- Victim Organization
- Cursor
- Incident Type
- Supply chain attack involving compromised credentials and distribution of modified Claude Code extension copies through the Cursor VS Code Extension Marketplace.
- Data Types Exposed
- Unredacted credentials
- Archived non-production source code
- Business contact information
- Customer support data
- Email conversations
- Product usage metrics
- Contact details
- Summaries of coding agent conversations
- Credentials users sent to the Cursor support team
- Affected Records
- Cursor said at most 5,000 users were affected; the number who installed or updated to a tampered extension version is not stated.
- Operational Impact
- Two modified versions of the anthropic.claude-code extension (2.1.312 and 2.1.313) were available through the Cursor VS Code Extension Marketplace for roughly 11 hours. Cursor said it blocked the attacker’s access after discovering the unauthorized activity on October 29 and considered the incident contained, with no attacker activity observed in its environment since November 1. The article does not specify the year for these dates.
- Ransom Or Extortion
- None reported for the Cursor incident.
- Claim Status
- confirmed
MITRE ATT&CK
T1078 · Valid AccountsThe article says internal Fireworks credentials were used to access stored secrets and account and billing information.T1195.002 · Compromise Software Supply ChainCursor described malicious, unauthorized copies of an extension being placed on the cache used to serve its extension marketplace, constituting a software supply chain compromise.
Vendors
Anthropicattack was discovered on October 29 and resulted in threat actors making two modified versions of the anthropic.claude-code extension (2.1.312 and 2.1.313) available through the Cursor VS Code Extension MarketplaceCursorCredentials compromised in the hack of AI inference platform Fireworks AI allowed attackers to steal more secrets and push malicious Claude Code extension copies, Cursor says.Fireworks AICredentials compromised in the hack of AI inference platform Fireworks AI allowed attackers to steal more secrets and push malicious Claude Code extension copies, Cursor says.
Products
Claude CodeCredentials compromised in the hack of AI inference platform Fireworks AI allowed attackers to steal more secrets and push malicious Claude Code extension copies, Cursor says.Cursor VS Code Extension Marketplaceof the anthropic.claude-code extension (2.1.312 and 2.1.313) available through the Cursor VS Code Extension Marketplace for roughly 11 hours.
