Kivorent News

Wednesday, November 18, 2026

Tech and AINews

Cursor says a key exposed by Fireworks AI led to its fake Claude Code extension

Cursor says the attacker who planted a malicious Claude Code extension in its marketplace began with an access key exposed by Fireworks AI, then used credentials found in old coding agent traces to reach internal tools.

Warden, covering security, outages and the rules for tech
, 3 min read

5,000most users who could have received the tampered build

What to know

  • Cursor says an attacker used an access key it had issued to Fireworks AI, which Fireworks exposed, to copy coding agent traces from a shared storage bucket.
  • Credentials in those traces opened some of Cursor's SaaS tools, and a stolen credential let the attacker plant tampered Anthropic.claude-code 2.1.312 and 2.1.313 builds on October 29.
  • Cursor has moved all inference off Fireworks, rotated its internal credentials and says the last attacker activity, on November 1, was blocked.

Cursor, the AI coding editor, says the attacker who slipped a malicious copy of Anthropic's Claude Code extension into its extension marketplace on October 29 got in through an access key that Cursor had issued to Fireworks AI, an inference provider it used. Cursor published the finding on November 13 after what it called two weeks of investigation with outside forensics experts.

How the attacker got in

According to Cursor's update, the key was exposed by Fireworks and let the attacker copy data from a read-only storage bucket the two companies shared. The bucket held coding agent traces, mostly from 2025, from sessions where users had not turned on Privacy Mode. A small number of those traces contained credentials that had not been redacted.

The attacker used those credentials to reach some of the software-as-a-service tools Cursor runs internally. From there, Cursor says, a stolen credential let the attacker place tampered copies of the Claude Code extension in the marketplace cache. Cursor does not say how Fireworks exposed the key.

Fireworks has disclosed two intrusions of its own in October, according to The Register: unauthorized use of internal credentials detected on October 6, and the compromise of a third-party cloud provider hosting its research machines on October 23. The Register reported on November 10 that Fireworks was working with customers including Cursor to assess the impact.

What was taken

Cursor lists archived, non-production source code repositories, limited customer support data and emails, product usage metrics and contact details, summaries of some coding agent conversations, and some credentials users had sent to its support team. It says coding agent traces were exposed for fewer than 0.05% of users and API credentials for less than 0.01% of customers, and that it contacted each affected customer directly. It does not give a total number.

Cursor says production systems, the Cursor app and its official updates, release signing keys, live customer traffic and customer code repositories were not affected.

The tampered extension

Versions 2.1.312 and 2.1.313 of the extension, for Windows and macOS, were live for about 11 hours on October 29, between about 2:27am and 1:35pm Pacific time. Cursor says at most 5,000 users who requested those cached versions may have received them. Anthropic did not publish them, Cursor says, and the upstream Open VSX registry never carried the malware.

Cursor's original notice says the payload collected host and user details, tried to add an attacker's SSH key labeled "svc-telemetry," opened a reverse SSH tunnel, and installed persistence that keeps running after the extension is removed. Cursor publishes check and cleanup scripts and tells affected users to rotate credentials reachable from the machine and consider rebuilding it.

Cursor says it has moved all inference off Fireworks and cut its access to shared data, rotated every exposed credential and, as a precaution, every internal credential, rebuilt the marketplace, and added secret scanning to its repositories and other internal data.

Why it matters

The breach shows how an AI tool's suppliers and its stored agent logs can become the way in. A key held by a vendor opened old agent traces, and secrets left in those traces opened the company that ships code to developers' machines. It follows a similar pattern to Railway's October breach and the GitHub CI secret stealer. Our explainers cover software supply chain attacks and what a post-mortem is.

Sources

AI agentsAnthropicInternet, outages and security

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.