Breaking3 min read

Cursor traces Claude Code extension attack to a key exposed by Fireworks AI

Cursor says an access key it issued to inference provider Fireworks AI was exposed and used to plant a backdoored Claude Code extension and copy historical agent traces. Cursor has moved all inference off Fireworks.

AE
Aidose Editorial
Nov 17, 2026

Cursor has published the findings of its two-week investigation into last month's tampered Claude Code extension, saying the attack began with an access key Cursor had issued to Fireworks AI. "The key was exposed by Fireworks," the company wrote in a November 13 update. The attacker used it to copy data from a read-only storage bucket shared with Fireworks and, combined with credentials found there, to reach other Cursor systems.

On October 29 the attacker placed malicious builds of the Anthropic.claude-code extension, versions 2.1.312 and 2.1.313, into the cache serving Cursor's VS Code extension marketplace. The builds were available for about 11 hours and attempted to add an attacker SSH key, open a reverse SSH tunnel and install persistence on macOS and Windows. Cursor says at most 5,000 users who requested those cached versions that day may have received them, and that the last attacker activity it observed was blocked on November 1. Cursor stresses that Anthropic did not publish the tampered copy and that it was not a vulnerability in Anthropic's product.

The data the attacker reached included coding agent traces, mostly from 2025, from sessions not in Privacy Mode, some containing unredacted credentials; archived non-production source repositories; limited customer support data and emails, including credentials users had sent to support; product usage metrics; and summaries of some agent conversations. Cursor says API credentials were exposed for less than 0.01% of customers and agent traces for fewer than 0.05% of users.

Cursor says Privacy Mode prompts and code, customer repositories, production infrastructure, live inference traffic, the Cursor app and its release signing keys were not affected. In response it moved all inference traffic off Fireworks, deactivated the Fireworks key, rotated exposed and internal credentials, rebuilt the extension marketplace, added secret scanning across its repositories and redeployed production infrastructure. It says it believes the incident is contained.

Anyone who installed either tampered version should follow Cursor's remediation steps rather than simply uninstalling the extension, which the company says may not be enough, and should rotate credentials reachable from the affected machine. The incident also shows how much sensitive developer data, including historical agent traces, can end up shared with an inference provider.

AE
Aidose Editorial
Nov 17, 2026 · 3 min read
Back to News