# Cursor: agent traces held the credentials that reached its marketplace

> Cursor traces its October marketplace compromise to a shared data bucket where stored coding agent traces still held live credentials.

- **Severity:** High: A working attack on a widely used AI product or ecosystem. A fix may exist. Patch or mitigate this week.
- **Category:** Incident, Supply chain
- **Published by:** Cursor
- **Disclosed:** 2026-11-13
- **Affects:** Cursor, Claude Code, Fireworks AI
- **Research by:** Cursor Team
- **Primary source:** https://cursor.com/blog/security-incident-update
- **Page:** https://theinjection.dev/items/cursor-agent-trace-credential-exposure/

Cursor says an access key it issued to Fireworks AI let an attacker read stored coding agent traces that still contained unredacted credentials, and that those credentials opened the SaaS tools used to tamper with its marketplace.

## What happened

Cursor published the root cause of its October 2026 marketplace incident on November 13, 2026, after about two weeks of investigation. The Cursor Team says an attacker obtained an access key that Cursor had issued to Fireworks AI, and that Fireworks exposed that key. Cursor says the last attacker activity it saw was on November 1 and was blocked.

## How it works

The key gave read access to a storage bucket Cursor used to share data with Fireworks. That bucket held coding agent traces from sessions outside Privacy Mode, mostly from 2025, and some of those traces still carried unredacted credentials. The attacker used the credentials to reach SaaS tools that Cursor uses, then used a stolen credential to place tampered Claude Code extension builds in the cache serving Cursor's marketplace.

## Who is affected

Cursor puts API credential exposure at under 0.01% of customers, and trace exposure at under 0.05% of users who were not in Privacy Mode. Archived non-production source repositories, limited support records and email, product usage metrics, contact details, conversation summaries and credentials users had sent to support were also reachable. Cursor states that Privacy Mode prompts and code, production systems, customer code repositories and its signed app updates were not affected.

## What to do

Rotate every credential named in Cursor's notification, which Cursor says it sent to affected admins as lists of specific keys, and rotate anything ever sent to Cursor support. Machines that installed Anthropic.claude-code 2.1.312 or 2.1.313 from the Cursor marketplace still need the steps in the October 29 notice, which Cursor warns uninstalling alone may not replace.

## Quick facts

- **Root cause:** An access key Cursor issued to Fireworks AI, exposed by Fireworks, per Cursor
- **Exposed:** Coding agent traces from sessions outside Privacy Mode, mostly from 2025
- **Credentials:** API credentials exposed for under 0.01% of customers, per Cursor
- **Traces:** Under 0.05% of users not in Privacy Mode, per Cursor
- **Tampered builds:** Anthropic.claude-code 2.1.312 and 2.1.313, available about 11 hours on October 29
- **Reach:** At most 5,000 users requested the cached builds, per Cursor
- **Status:** Fireworks key deactivated, inference moved off Fireworks, credentials rotated, marketplace rebuilt

## Questions

### Was my Cursor data in the exposed coding agent traces?

Cursor says it contacted affected customers directly, and that anyone who received no notification was not in the exposed traces. Exposure was limited to sessions outside Privacy Mode, which Cursor puts at under 0.05% of those users, with API credentials exposed for under 0.01% of customers. Cursor directs questions to security@cursor.com.

### Which credentials should I rotate after the Cursor incident?

Rotate the specific keys listed in the notification Cursor sent to affected admins, and rotate any credential ever sent to Cursor support, because support records and email were reachable. Credentials that appeared inside coding agent sessions outside Privacy Mode also count, since Cursor says some stored traces kept them unredacted.

### Did the Cursor incident come from a flaw in Anthropic's Claude Code?

No. Cursor states the incident was not a vulnerability in Anthropic's product, and that the attacker published the tampered builds using a stolen Cursor credential. Cursor adds that Open VSX never contained the malware, and that its own app and official updates were not signed with exposed keys. The affected builds were Anthropic.claude-code 2.1.312 and 2.1.313.

### What does Cursor's November 13 update add to its October 29 notice?

Cursor's October 29 notice covered the tampered Claude Code builds, indicators and remediation. The November 13 update names the entry point, a shared Fireworks bucket of stored agent traces rather than Cursor's production systems, and widens the known data exposure to archived non-production repositories, support records, usage metrics and conversation summaries.

## Sources

1. [Cursor Incident Update](https://cursor.com/blog/security-incident-update)
2. [Cursor Security Notice](https://cursor.com/blog/claude-code-extension-security-notice)
3. [Fireworks Bulletin](https://fireworks.ai/security/bulletins/october-2026-security-incident-update)
