Cursor Traces Marketplace Backdoor to Exposed Fireworks Key
Cursor's post-incident report, published November 13, says its October breach started with an access key it had issued to Fireworks AI, its inference vendor. The attacker copied coding-agent traces from a shared storage bucket, found unredacted credentials inside a small number of them, and used one on October 29 to place tampered copies of Anthropic's Claude Code extension on the cache serving Cursor's marketplace. At most 5,000 users received the malicious builds, which tried to plant an SSH backdoor on developer machines.
The break-in didn't start at the marketplace. It started in a read-only bucket of old coding sessions, where a handful of unredacted credentials turned a vendor's exposed key into the keys to the store.
Cursor has traced its October security breach to an access key it issued to Fireworks AI, its third-party training and inference vendor. In a post-incident report published on Thursday, November 13, the company said the key was exposed by Fireworks, and an attacker used it to copy data from a read-only storage bucket Cursor used to share data with the vendor.
The bucket held coding-agent traces, mostly from 2025, drawn from sessions that weren't in Privacy Mode and that Cursor was permitted to use for training. A small number of those traces contained credentials nobody had redacted. Cursor says fewer than 0.05 percent of non-Privacy Mode users had traces exposed, and fewer than 0.01 percent of customers had API credentials exposed.
One of those credentials opened the door that mattered. On October 29, the attacker used it to write tampered builds of Anthropic's Claude Code extension — versions 2.1.312 and 2.1.313 — directly onto the cache that serves Cursor's extension marketplace. The malicious versions were served between 09:27 and 20:35 UTC, about eleven hours, before Cursor removed them and rebuilt the marketplace.
Cursor says at most 5,000 users who requested those cached versions potentially received the modified package. Anthropic didn't publish the tampered builds, and the compromise wasn't a vulnerability in Anthropic's product. Open VSX, the upstream marketplace source, never carried the malware, and Microsoft Threat Intelligence reports the packages never appeared on the Visual Studio Marketplace or in the npm package either.
The tampered extension collected host and user details — hostname, username, operating system, home directory path, environment variable names, working directories, the configured Git email address and the machine's public IP address. It then tried to add an attacker-controlled SSH key, labeled svc-telemetry, to the machine's authorized_keys file, switch on the local SSH service and open a reverse SSH tunnel back out.
That last step is where the attack mostly failed. Turning on the SSH service takes administrator rights, which the extension doesn't have on most macOS and Windows machines, so the tunnel never opened on those hosts. Security analysts still advise anyone who installed the affected versions to check for the planted key and the dropped files, because a failed tunnel doesn't remove what the extension left behind.
Cursor says the last attacker activity it saw came on November 1 and was blocked. The company has moved all of its inference traffic off Fireworks, deactivated the vendor's key, rotated every credential used across its internal systems and added secret scanning to its repositories and other internal data stores. It says Privacy Mode prompts and code, customer code repositories, its production systems and the Cursor app itself weren't affected.
The incident reaches past one vendor relationship. Fireworks disclosed several breaches of its own in October, one involving exposed API keys and tokens, and Notion has confirmed data exposure through the same vendor. Uber's California breach filing describes a service provider whose credentials were compromised, and reporting by TechCrunch identified that provider as Fireworks. A single vendor's exposed key became the first link in a chain that ended inside thousands of developers' editors — and the credentials that carried the attacker from link to link were sitting, unredacted, in data that was only ever meant to be read.
Sources
- Trojanized anthropic.claude-code 2.1.313 in the Cursor marketplace planted an SSH backdoor on up to 5,000 developer machines — Corgea research, updated November 15, 2026
- Cursor Security Breach: Fireworks Key and Fake Claude Code Extension — iFynx, November 14, 2026
- The tampered Claude Code extension in Cursor's marketplace: what teams should check — ReclaimLLM, updated November 14, 2026
- Cursor Links Claude Code Extension Attack to Fireworks AI Breach — Windows Forum, November 2026
The conversation.
Loading comments…
A little more perspective.
Reading is free. Verified members can comment and reply for $2 a month.