Cursor acknowledges breach after hackers infiltrate marketplace to distribute malicious Claude Code add‑on

On Monday, Cursor—an AI‑assisted coding platform—revealed that it had been compromised, enabling attackers to post a dangerous Claude Code extension in its public marketplace. The firm traced the breach to stolen credentials that granted the intruders illicit entry to its extension‑publishing system.

Investigators found that the hijacked credentials came from Fireworks AI, the AI infrastructure supplier for Cursor. Using those Fireworks AI login details, the perpetrators posed as authorized developers and uploaded the malicious package, bypassing the platform’s usual review procedures.

The counterfeit Claude Code extension was built to run concealed code on any machine where it was installed, opening the door to data theft, ransomware attacks, or additional lateral movement inside corporate networks. Since the add‑on was listed in Cursor’s official marketplace, numerous developers downloaded it believing it had cleared security vetting.

After detecting the intrusion, Cursor swiftly took down the extension, regenerated all internal access tokens, and released a notice advising users to uninstall the package and scan their systems for any anomalous behavior. The firm also said it will strengthen authentication controls and tighten the vetting process for third‑party extensions.

The incident joins an expanding tally of supply‑chain assaults on AI development tools, where threat actors leverage the trust inherent in shared repositories to reach large developer audiences. Comparable methods have surfaced in other software ecosystems, underscoring the necessity for stronger verification and monitoring of code contributions.

Law‑enforcement agencies and cybersecurity companies have been enlisted to probe the origin of the Fireworks AI credential compromise, and Cursor signaled it may seek legal recourse against the attackers. Specialists advise developers to implement multi‑factor authentication, routinely audit third‑party dependencies, and remain vigilant for abnormal behavior in newly installed tools as the sector wrestles with the security challenges of a fast‑growing AI‑driven software landscape.