Account Security

Malicious Claude Code Extension Hit Up to 5,000 Cursor Users

Cursor says attackers distributed two malicious Claude Code extension versions that could install persistent SSH access on computers.

Malicious Claude Code Extension Hit Up to 5,000 Cursor Users — illustration

Cursor says attackers used a stolen marketplace credential to distribute malicious copies of the Claude Code extension to as many as 5,000 users. Anyone who uses Cursor for personal projects or a small business should check the installed extension version and replace credentials that may have appeared in AI coding sessions.

What happened

Attackers first obtained an access key held by Fireworks AI, a third-party service Cursor used for AI training and inference, according to BleepingComputer. That key provided access to a read-only storage bucket containing historical Cursor coding-agent traces, mostly from 2025.

Some traces included API credentials that had not been removed. The attackers reportedly used those credentials to enter additional Cursor systems and steal information from several software services. Cursor says fewer than 0.05% of users had traces exposed and fewer than 0.01% of customers had API credentials compromised.

The attackers then used a stolen credential to place unauthorized Claude Code extension builds in Cursor’s marketplace on October 29. Versions 2.1.312 and 2.1.313 were available for about 11 hours before removal.

BleepingComputer reports that the altered extension collected system details, added an attacker-controlled SSH key, tried to enable SSH access and created ways to retain access after a restart. Coding mistakes may have stopped the backdoor on many Windows and Mac computers where SSH was not already active, but affected users should not assume their machine was protected.

Anthropic’s legitimate extension and the Open VSX marketplace were not compromised, according to the report. Cursor also says its production systems, official app releases and customer code repositories were not breached.

Who is affected

Cursor estimates that up to 5,000 people who requested either malicious extension version during the exposure window may have received it. Developers using Cursor at home, for freelance work or in a small office should check, even if they do not remember installing an update manually.

The exposed coding traces came from users who had not turned on Privacy Mode and whose information could be used for model training. Cursor says code and prompts belonging to Privacy Mode users were not affected. The source does not disclose a complete list of compromised customers or services.

What to do now

  1. Open Cursor’s Extensions view, find Claude Code and check its version. If it shows 2.1.312 or 2.1.313, stop using the computer for sensitive work and remove the extension.

  2. Have a qualified IT or security professional inspect an affected computer for an unfamiliar SSH key, enabled SSH access and persistence changes. The malicious code specifically attempted all three actions.

  3. Review any passwords, API keys or service tokens that may have appeared in Cursor prompts or coding-agent traces. Sign in directly to each affected service, open its security or API-key settings, revoke the old credential and create a replacement.

  4. Replace reused passwords with unique ones and store them securely. This limits the damage if one exposed login is tried against other accounts.

A password manager makes unique credentials easier to create and maintain across a household or small office.

Bitwarden Premium · 9/10 · list $19.80 See today's price at bitwarden.com for Bitwarden Premium

  1. Open Cursor’s privacy settings and enable Privacy Mode if you do not want code and prompts included in model-training data. Cursor says information from users with that mode enabled was not part of this exposure.

  2. Watch important developer and cloud accounts for unfamiliar sessions, new API keys or unexpected configuration changes. Revoke anything you do not recognize and contact the service provider if access cannot be explained.

Sources

Written by the CyberSec24 news desk with AI assistance, from the sources above. Spot an error? Tell us and we'll correct it.