Fireworks AI breach triggers Cursor supply chain attack
A compromised cloud provider for Fireworks AI led to a supply chain attack, with stolen credentials used to distribute malicious versions of the Anthropic Claude Code extension via Cursor's marketplace to up to 5,000 users.

A breach at Fireworks AI's cloud provider on October 23-24, 2026, enabled a supply chain attack that distributed malicious developer tools through the Cursor marketplace. Attackers used stolen internal Fireworks credentials to access a Cursor access key and place tampered extensions in the platform's cache.
Fireworks AI reported unauthorized access to its development machines on October 24, with no further successful access detected after October 25. The attacker used compromised credentials to access stored secrets and account information. A Cursor access key issued to Fireworks was stolen, allowing the attacker to copy information from a read-only storage bucket shared with the vendor. This bucket held limited coding-agent traces from 2025, some of which retained credentials that had escaped redaction.
Malicious extension distribution via Cursor cache
Attackers placed unauthorized copies of Anthropic's Claude Code extension directly into the cache serving Cursor's VS Code extension marketplace on October 29, 2026. The malicious activity occurred in Cursor's environment, not in the Open VSX marketplace source. The malware was served to users who requested cached versions of the extension during an 11-hour window.
The modified extension collected host information and attempted to establish a persistent SSH backdoor, which generally failed due to lack of administrator privileges. The incident involved two specific versions of the extension for macOS and Windows.
| Version | Release Date |
|---|---|
| 2.1.312 | October 29, 2026 |
| 2.1.313 | October 29, 2026 |
The upper bound on potential delivery was approximately 11 hours, from 2:27 a.m. to 1:35 p.m. Pacific time on October 29. This is not a count of successful installations.
Scope of impact and response
At most 5,000 users were affected by the availability of the malicious extension. Cursor disclosed the breach on November 13, 2026, after a two-week investigation. The company considers the incident contained, with the last observed attacker activity on November 1.
Cursor contacted affected customers and sent lists of exposed keys for rotation. The company engaged external forensic experts. The attacker accessed select archived non-production Cursor source code repositories, some of which contained business contact information. Limited customer support data and email conversations were accessed, along with product usage metrics and contact details. Summaries of some coding agent conversations were accessed, as were some credentials users had sent to the Cursor support team.
Third-party reporting and vendor statements
Media outlets reported the breach chain on November 14 and 15, 2026. Anthropic confirmed it did not build or publish the tampered copy and stated its official channels were unaffected. The malicious activity was entirely within Cursor's systems following the initial Fireworks AI compromise.
The credentials enabled access to selected SaaS tools used by Cursor. Cursor blocked the attacker's access immediately after discovering the unauthorized activity on October 29.





