Cursor Breach Began With a Key Issued to Fireworks and Reached Credentials Users Sent to Support

What happened
Cursor says an attacker used a Cursor-issued key exposed by Fireworks to copy mostly 2025 traces from sessions outside Privacy Mode, then reached source code, support data and credentials sent to support.
Why it matters
Secrets users pasted into support threads and API keys in traces may be in the attacker's hands; Cursor last saw attacker activity on 1 November.
What to do
Rotate any credential you sent Cursor support, rotate keys on any admin list Cursor emailed, and follow Cursor's security notice in full if you installed extension 2.1.312 or 2.1.313.

Rotate any API key, token or password you ever pasted into a message to Cursor support, today, whether or not Cursor has emailed you. Cursor's own update says the attacker reached "some credentials users had sent to the Cursor support team", and a secret that sat in a support thread should now be treated as public. Our Cursor verdict stays conditional, and the same incident is behind the trojanized Claude Code extension that appeared in Cursor's marketplace on 29 October. Fireworks AI, the inference provider whose exposure started it, is not in our directory.

What happened

Cursor's update, published 13 November 2026, says: "An attacker obtained an access key we had issued to Fireworks. The key was exposed by Fireworks, and the attacker used it to copy data from a read-only storage bucket." That bucket held coding agent traces "mostly from 2025, from sessions not in Privacy Mode".

  • Traces: Cursor puts the trace exposure at fewer than 0.05% of users.
  • API credentials: "less than 0.01% of customers had API credentials exposed". Team admins were emailed a list of the specific keys to rotate.
  • Beyond the bucket: the attacker reached "select archived non-production Cursor source code repositories", "limited customer support data and email conversations", summaries of some coding agent conversations, and the support credentials above.
  • Timeline: "The last attacker activity we have seen was on November 1."
  • Response: Cursor says it "Moved all inference traffic off Fireworks" and deactivated the key it had issued.

All of these figures are Cursor's own; no independent party has verified them.

The malicious extension

This is the incident behind the trojanized Anthropic.claude-code extension, versions 2.1.312 and 2.1.313, in Cursor's marketplace on 29 October. Cursor now says "At most 5,000 users who requested specific cached versions of the extension on October 29 potentially received the malicious build", which was available for approximately 11 hours. Its warning stands: "Uninstalling the extension alone may not be enough." The problem was the build served through Cursor's marketplace, not Claude Code itself.

Fireworks and Uber

The Hacker News, reporting on 15 November, adds that Uber disclosed in a California filing that a service provider notified it on 9 October of compromised credentials. Uber did not name the provider. That it was Fireworks is TechCrunch's reporting as cited by The Hacker News, not Uber's own words.

Why it matters

The attack path is the lesson. A key Cursor handed to an inference provider gave an outsider read access to user traces, and from there to support data that held live secrets. The deciding setting was Privacy Mode: the exposed traces came from sessions without it, which Cursor says it was permitted to use for model training.

Cursor's update is unusually specific, which is to its credit, and it is why the verdict holds at conditional rather than dropping. But every hosted coding agent sends your code through a chain of vendors you never see, and this is what happens when one link leaks.

What changes for you

  • Support credentials: rotate anything you ever sent to Cursor support, even without a notice.
  • Team admins: rotate every key on the list Cursor emailed you.
  • Extension 2.1.312 or 2.1.313: if you installed it from Cursor's marketplace on 29 October, follow Cursor's security notice in full, including its indicators of compromise and check scripts. Uninstalling is not enough.
  • Privacy Mode: turn it on, or its equivalent in any hosted coding agent, for sessions that touch sensitive code.

What to do

  1. 1 Rotate any credential you ever sent to Cursor support, even without a notice.
  2. 2 Team admins: rotate the keys on the list Cursor emailed you.
  3. 3 If you installed Anthropic.claude-code 2.1.312 or 2.1.313 from Cursor's marketplace on 29 October, follow Cursor's security notice and its check scripts in full; uninstalling is not enough.
  4. 4 Turn on Privacy Mode for sessions that touch sensitive code.

Affected tools & models

CursorClaude CodeFireworks AI

Never need to catch up again

The weekly delta: only verdict changes and act-now items. No digest filler.

By subscribing you agree to our Privacy Policy. Unsubscribe anytime.