Cursor confirms breach behind malicious Claude Code extension
The code editor company said attackers used credentials stolen from AI infrastructure provider Fireworks AI to push a malicious extension through its marketplace.
By The Agentic Times ·

Cursor has confirmed a security breach that allowed attackers to distribute a malicious Claude Code extension through its marketplace, according to a report published by BleepingComputer on 14 November 2026.
The company said the attackers obtained credentials stolen from Fireworks AI, an AI infrastructure provider, and used them to publish the extension. Claude Code is a coding agent, meaning software that can read, write and run code on a developer's behalf. Extensions distributed through an editor marketplace typically run with the same access as the editor itself, which is why marketplace compromises are treated as supply chain incidents rather than isolated bugs.
The root cause, as described in the confirmed account, was the supply chain rather than a flaw in Cursor's own code. Stolen credentials belonging to a third party were enough to get attacker-controlled software in front of users through a trusted distribution channel. That pattern removes the usual warning signs for developers, who have no obvious way to tell a tampered extension from a legitimate one at the point of installation.
The reporting does not state how many users installed the malicious extension, how long it remained available, or what the extension did once installed. It also does not say when the compromise of Fireworks AI credentials first occurred, how those credentials were obtained, or whether other marketplaces or vendors were affected by the same set of credentials. Cursor has not, in the material available, published a breakdown of affected accounts or a timeline of its own detection and response.
For developers, the practical question is whether a Claude Code extension installed from the Cursor marketplace came from the legitimate publisher. Until the vendor publishes version numbers, publishing dates or indicators of compromise, that question cannot be answered from the outside. Teams that manage developer tooling centrally may want to record which machines have the extension installed, so that a later advisory can be acted on quickly.
The incident is the kind that security teams have been warning about as coding agents spread through development environments. An agent extension is not a passive plugin. It can hold API keys, read repository contents and execute commands, so a compromised one inherits a wide blast radius inside an organisation's source code.
Cursor has confirmed the breach. Further detail on scope and remediation will depend on what the company and Fireworks AI publish next. The Agentic Times will update this report if either issues a fuller account.
Sources
- bleepingcomputer.comhttps://bleepingcomputer.com/news/security/cursor-confirms-security-breach-led-to-malicious-claude-code-extension
