Answers · Cursor

Cursor's security incident: what leaked, who's affected, what to rotate

If Cursor didn't contact you, its Nov 13 update says your data wasn't in the exposed traces: "If you did not receive a notification, your data was not in these traces." Privacy Mode prompts and code weren't affected, and Cursor says it "contacted each affected customer directly." The one thing to check yourself: whether you installed Anthropic.claude-code 2.1.312 or 2.1.313 from Cursor's extension marketplace on Oct 29.

If Cursor did contact you: rotate every key on the list it emailed your team admin, at the provider that issued it, and confirm each old value is dead. If you got the tampered extension, uninstalling isn't the fix: "Uninstalling the extension alone may not be enough." Run Cursor's check below, then rotate whatever that machine could reach, from a different machine.

What happened, per Cursor

Not affected, per Cursor: Privacy Mode prompts and code, live traffic, production systems, the Cursor app and its official updates ("no release signing keys were exposed"), every other marketplace extension, plugins from the Cursor Marketplace, active inference requests, and customer code repositories. Cursor has moved all inference off Fireworks. Fireworks is the same inference vendor behind Notion's Nov 9 notice: Pasted a key into Notion AI?

  • It started at Fireworks. "An attacker obtained an access key we had issued to Fireworks. The key was exposed by Fireworks." With it the attacker copied a read-only storage bucket Cursor used to share data with Fireworks: "limited coding agent traces, mostly from 2025, from sessions not in Privacy Mode, which we were permitted to use for model training."
  • Some traces held live credentials. "A small number of those traces contained credentials that had not been redacted; less than 0.01% of customers had API credentials exposed." Agent traces were exposed for "fewer than 0.05%" of non-Privacy-Mode users.
  • Those credentials opened Cursor's SaaS tools. From there: archived non-production source repositories, "limited customer support data and email conversations, product usage metrics and contact details, summaries of some coding agent conversations, and some credentials users had sent to the Cursor support team."
  • Oct 29: a tampered extension. Using "a stolen credential from the exposed data," the attacker put malicious copies of Anthropic's Claude Code extension on the cache that serves Cursor's VS Code extension marketplace. Up to 5,000 users who requested those cached versions "potentially received the malicious build"; it was up about 11 hours (2:27am to 1:35pm PT, per Cursor's Oct 29 notice). Anthropic didn't publish it, and Open VSX, the marketplace source, "never had the malware."
  • From Oct 30: notice. Cursor published indicators of compromise and remediation scripts and began notifying affected users and team admins. Where credentials were exposed, it emailed team admins "a list of the specific keys to rotate," and asked the services that issued those keys to revoke them.
  • Nov 1: last activity. "The last attacker activity we have seen was on November 1, and it was blocked." Cursor believes the incident is contained.

Check for the tampered extension (macOS, Windows)

Cursor's notice covers Anthropic.claude-code 2.1.312 and 2.1.313 for macOS and Windows, installed or updated from the Cursor VS Code Extension Marketplace between 2:27am and 1:35pm PT on Oct 29, 2026. The payload adds an SSH key labeled svc-telemetry, installs persistence, and sends host details (hostname, username, environment variable names, Git email, public IP) to the attacker. Cursor says the attacker's server no longer works.

  1. Cursor's check on macOS. If the key is there, the machine is affected:

    grep -qs svc-telemetry ~/.ssh/authorized_keys && echo "svc-telemetry key found" || echo "No key found; machine not impacted"
  2. Cursor's check on Windows (PowerShell):

    $files = "$HOME\.ssh\authorized_keys","C:\ProgramData\ssh\administrators_authorized_keys" | Where-Object { Test-Path $_ }
    if ($files -and (Select-String svc-telemetry $files)) { "svc-telemetry key found" } else { "No key found; machine not impacted" }
  3. Our extra look on macOS, using the files and processes in Cursor's indicator list. Pass: no output from any line; investigate any hit:

    ls -la ~/.config/.smt ~/.config/.smtk ~/Library/LaunchAgents/com.apple.sysmond.plist 2>/dev/null
    ls -a "$TMPDIR" | grep '^\.stk_'
    pgrep -fl 'ssh -N -R|autossh -M 0'

    On Windows the indicators include a scheduled task named SysMon (schtasks /query /tn SysMon), %USERPROFILE%\.config\smt.cmd and a Startup-folder sysmon.cmd.

  4. Found something? Follow Cursor's steps: isolate the host, run its remediation script (remediate-macos.sh as the affected user, not root; remediate-windows.ps1 as Administrator), and decide whether to rebuild. Then "rotate credentials and sessions accessible from affected hosts, from a trusted system." Cursor: "The limited env-var collection does not rule out later credential access over the backdoor."

If Cursor emailed you a key list: rotate and prove it

  1. For each key on the list: create the replacement at the provider, update every place that uses it, then revoke the old value. Cursor asked the issuers to revoke these keys, so some may already be dead; don't assume which.

  2. Prove each old value fails. Pass: 401 (or 403). Our commands; put the old value in OLD first:

    curl -s -o /dev/null -w '%{http_code}\n' https://api.openai.com/v1/models -H "Authorization: Bearer $OLD"
    curl -s -o /dev/null -w '%{http_code}\n' https://api.anthropic.com/v1/models -H "x-api-key: $OLD" -H 'anthropic-version: 2023-06-01'
    curl -s -o /dev/null -w '%{http_code}\n' https://api.stripe.com/v1/balance -u "$OLD:"
    curl -s -o /dev/null -w '%{http_code}\n' https://api.github.com/user -H "Authorization: Bearer $OLD"
  3. Read each provider's usage and audit logs for activity you don't recognize. Cursor's update doesn't date the bucket copy, so look back as far as the logs go, through at least Nov 1.

  4. Ever pasted a credential into a Cursor support email? Cursor says "some credentials users had sent to the Cursor support team" were accessed and that it notified affected customers directly. Our note: if you're not sure, rotating that one key costs ten minutes.

What went wrong, and the version of it in your own app

  • Secrets in agent transcripts. The root was credentials sitting unredacted in coding agent traces. Our note: an agent session that runs cat .env or echoes a token writes it into a log someone keeps. Give agents placeholder values, and scan your own logs and transcripts for key prefixes (sk-, sk_live_, AKIA, ghp_, github_pat_).
  • A long-lived key held by a vendor. Cursor now restricts provider credentials "so compromised credentials can't be used outside their intended location." Our note: IP or network allowlists turn a leaked key into a useless string.
  • Support inboxes keep what you send. Credentials users mailed to support were still sitting in Cursor's support data when the attacker reached it. Never send a live key to any vendor's support; send a fresh, scoped, short-lived one and revoke it after.
  • One credential led to the next. Bucket key, then trace credentials, then SaaS tools, then the marketplace cache. Cursor's fix list: rotate every credential "even where we had no evidence of exposure," plus secret scanning across all repositories and other internal data sources.

Ship Check Kit · $29

Rather not run it by hand?

The Ship Check Kit can't tell you whether you were in Cursor's dataset or whether your laptop has the backdoor; use Cursor's check above for that. What it does, run by your coding agent on your app: SC-SEC-01 and SC-SEC-02 find keys that left the server, SC-SEC-03 fails while any old value still gets anything but a 401 or 403, SC-SEC-04 flags secrets written to logs and outbound payloads, and SC-AI-02 fails any AI provider without a spend ceiling and alert. Five of 44 launch checks, each with the evidence and the fix.

Checked against Cursor: An update on our recent security incident (Nov 13), Cursor: Security notice, malicious Anthropic.claude-code 2.1.312 and 2.1.313 (Oct 29, IOCs updated Oct 30), Fireworks AI: Security bulletins on 2026-11-16. Only test apps and accounts you own or are allowed to test.