Jones Web Designs & Development is now The Jones Foundry. What it means for clients

Cursor breach sent malicious Claude Code extension to developers

A stolen access key exposed Cursor data and let attackers publish trojanized versions of Anthropic's Claude Code extension to its VS Code marketplace, potentially affecting thousands of developers.

Cursor users and developers were exposed after attackers stole an access key stored at Fireworks AI and used it to copy data from a shared read-only bucket, according to BleepingComputer. That data included historical AI coding-agent traces, some containing unredacted API credentials, and the attackers used exposed credentials to access internal Cursor systems and customer data.

Cursor breach: how the extension was trojanized

According to BleepingComputer, the attackers used credentials obtained from the exposed traces to reach internal services and, ultimately, a credential that allowed them to place malicious copies of Anthropic's Claude Code extension on Cursor's VS Code extension marketplace. The malicious builds, identified as versions 2.1.312 and 2.1.313, were available for roughly 11 hours before Cursor removed them. Cursor estimates that at most 5,000 users who requested those cached versions during the exposure window may have received the malicious builds. The Hacker News reports the same chain of events, noting the initial theft of the Fireworks-issued key and the subsequent use of data pulled from the shared bucket.

The injected code attempted to install persistent SSH backdoors. BleepingComputer reports the malware gathered system details, tried to add an attacker-controlled SSH key, and attempted to enable SSH access with persistence mechanisms. Security researchers found coding errors that probably stopped the backdoor from working on many Windows and macOS systems where SSH was not already enabled. Cursor says production infrastructure, official app releases, customer code repositories, and data from users who had enabled Privacy Mode were not affected.

What data and systems were accessed

Cursor told investigators that attackers copied material from a read-only storage bucket used to share information with Fireworks AI, and some of the copied traces dated mostly from 2025. The traces came from users who had not enabled Cursor's Privacy Mode and had been allowed for model training. Fewer than 0.05% of users had coding-agent traces exposed, and fewer than 0.01% of customers had API credentials compromised, according to Cursor as reported by BleepingComputer.

Using exposed credentials, attackers accessed a set of Cursor SaaS tools and took limited archived non-production source code repositories, some containing business contact information, as well as customer support emails, product usage metrics, summaries of some coding-agent conversations, and credentials users had previously shared with Cursor support. Cursor says its monitoring detected unauthorized activity, which led it to revoke the Fireworks access key, rotate exposed credentials, and investigate further.

Why this matters to small teams and nonprofits

Supply chain and developer tool compromises can reach far beyond a single product. In this case, attackers abused a third-party vendor key to get to developer traces that contained credentials. Those credentials then let attackers publish malicious extension builds that could install remote access on developer machines. BleepingComputer places this incident in a wider wave of attacks targeting AI development platforms, sometimes called "Black October," where stolen keys and tokens were used for unauthorized AI inference and other activity.

For small organizations, the practical risk is twofold: stolen credentials can let attackers move laterally into other services, and compromised developer tools or extensions can deliver malware to machines that run those tools. Even when a live backdoor fails because of coding errors or disabled services, the attempt still indicates attackers can reach build and distribution systems.

Practical steps to reduce your risk

  • Check installed VS Code extensions and update or remove any that were fetched or updated on October 29, 2026, if you used Cursor's marketplace during that time, and verify versions with your vendor. (Cursor identified versions 2.1.312 and 2.1.313 as malicious.)
  • Revoke and rotate any API keys or credentials that may have been shared with third-party tooling, and avoid storing long-lived keys in code, agent traces, or support tickets.
  • Enable vendor privacy features where available to opt out of training data collection, and review how your tools store agent traces and logs.
  • Inspect authorized SSH keys on developer machines and servers, remove unknown keys, and check for unexpected persistence mechanisms.
  • Review access logs for suspicious activity, enable multi-factor authentication on developer and cloud accounts, and limit who can publish or sign extensions and packages.
  • If you use third-party AI training or inference vendors, ask them about key storage practices and incident history, and consider segregating data sets used for training from production secrets.

These steps follow directly from the facts reported by Cursor and BleepingComputer about how credentials and traces were used, and they point to low-cost checks most small teams can perform quickly.

What Cursor did and what remains uncertain

Cursor says it revoked the Fireworks access key, rotated exposed credentials, moved inference traffic away from Fireworks, rebuilt its extension marketplace, and redeployed production infrastructure as a precaution, according to BleepingComputer. Cursor also temporarily disabled some internal and product services while investigating. The company notified affected users and worked to help revoke exposed credentials. The Hacker News adds that Cursor said no additional malicious activity was detected after a given point during the investigation.

What remains unclear from the public reporting is the full list of compromised downstream services and whether any exposed credentials were used beyond the published actions. BleepingComputer notes the incident is part of a broader series of compromises affecting AI developer platforms that relied on shared keys and third-party services.

If you want to check how exposed your own site is to this kind of supply chain risk, start with a free external scan on our security page: /security.html. For more technical detail, read the original reporting from BleepingComputer and The Hacker News: BleepingComputer report and The Hacker News summary.

Sources

This post was drafted with AI from the reporting linked above and published by The Jones Foundry. For full details, read the original sources.

Found this useful? Pass it on to someone who would want to know.

All Online Security posts →