Cursor links malicious Claude Code extension to stolen Fireworks AI credentials
Cursor has confirmed that a security breach enabled attackers to distribute a malicious Claude Code extension through its marketplace. The company said the attackers obtained credentials from AI infrastructure provider Fireworks AI, connecting the credential theft to the extension’s distribution.
The disclosure identifies two enterprise concerns: the compromise of credentials held by a service provider and the use of a software marketplace to deliver a malicious extension. The available information does not specify when the breach occurred, how the credentials were stolen, how many users downloaded the extension, or what the extension did after installation. It also does not identify affected versions, provide indicators of compromise, or state whether a fix or removal has been completed. Organizations should avoid assuming those details until they are confirmed.
Why the incident matters to vendor risk
For organizations using developer tools and AI infrastructure, the incident is a reminder that supplier relationships can create interconnected exposure. Credentials associated with a provider may provide access to systems or services beyond the provider itself, while a marketplace can act as a distribution channel that brings software from outside an organization’s normal release process into developer environments.
This does not establish that any particular customer environment or Fireworks AI service was accessed. It does, however, make credential scope and extension governance relevant questions for security and procurement teams. Organizations should determine whether Cursor, Claude Code, or Fireworks AI are used in their environments, including through individual developer accounts or teams that may not be centrally managed.
Practical steps for security teams
-
Ask Cursor and Fireworks AI for incident details relevant to the organization, including the affected credentials or services, the exposure period, containment actions, and any customer-specific impact. Record what is confirmed separately from what remains unknown.
-
Review whether the Claude Code extension was installed or used by employees. Where telemetry is available, examine extension inventories and installation records, and follow vendor guidance before taking remediation actions.
-
Assess credentials and integrations connected to the relevant tools. If an organization identifies credentials that may be in scope, coordinate rotation or revocation with the responsible provider and internal owners, then check for unexpected use. The source information does not establish which credentials customers should change.
-
Review controls for developer extensions and marketplace software. Consider whether installation is limited to approved items, whether extensions are inventoried, and whether developers have a clear process for reporting unexpected software.
-
Update vendor-risk records with the confirmed relationship between the credential theft and the marketplace distribution, while avoiding unsupported conclusions about impact or attacker activity.
Information still needed
The public details provided so far are brief. A fuller assessment depends on answers about the stolen credentials’ permissions, the extension’s behavior, the number and identity of affected users, and the timeline for detection and containment. Those facts would help organizations distinguish a marketplace exposure from any downstream access to their own environments.
Until additional details are available, enterprises should treat the incident as a reason to verify usage, review extension controls, and seek direct guidance from the vendors involved. Cursor’s confirmation establishes that the breach enabled distribution of a malicious Claude Code extension and that credentials stolen from Fireworks AI were involved; it does not, by itself, establish the scope of customer impact.