It started with a single access key. On 13 November 2026, Cursor published an investigation update explaining how its October security incident unfolded. A key issued to a vendor was exposed, a storage bucket was copied, and a tampered copy of Anthropic's Claude Code extension ended up in front of as many as 5,000 users.
Cursor is one of the most widely used AI coding tools, so the details matter to developers well beyond its own user base. Here is what happened, who was affected, and what to check.
What happened, step by step
The story starts at Fireworks AI, one of Cursor's inference vendors. Cursor had issued Fireworks an access key. In Cursor's words, that key was "exposed by Fireworks". An attacker obtained it and used it to copy a read-only storage bucket Cursor used to share data with Fireworks.
That bucket held coding-agent traces — records of what Cursor's coding agent did during past sessions, mostly from 2025, from sessions not in Privacy Mode. Most were ordinary working records, but a small number contained credentials that had never been redacted.
Using credentials from the traces, the attacker then reached some of Cursor's internal software tools. On 29 October, it placed malicious copies of Anthropic's Claude Code extension — versions 2.1.312 and 2.1.313 — on the cache serving Cursor's VS Code extension marketplace.
The malicious extension
The tampered builds sat on the marketplace cache for about 11 hours, from 2:27 a.m. to 1:35 p.m. Pacific Time on 29 October. Cursor says at most 5,000 users who requested those cached versions received the malicious build — a worst case, not a confirmed count.
Two points limit the damage. Cursor says Anthropic did not publish the modified copies — the attacker placed them on Cursor's cache, not on Anthropic's official channel. The malware also never appeared on the upstream Open VSX registry, so the poisoned builds reached users only through Cursor's own marketplace cache.
Cursor disclosed the tampered extension on 29 October. The 13 November update is the first time it has publicly connected the dots back to the Fireworks key.
Who was affected
Cursor puts the exposure in two numbers. Coding-agent traces were exposed for under 0.05 percent of users — those whose sessions were outside Privacy Mode. API credentials were exposed for under 0.01 percent of customers. Small fractions, but Cursor has millions of users.
Cursor says Privacy Mode prompts and code, code repositories, production systems and the Cursor app itself were not affected. The last attacker activity it has seen was on 1 November.
What Cursor did about it
Cursor revoked the exposed key, moved all inference traffic off Fireworks, and cut off the shared data arrangement. It rotated the exposed credentials first, then all internal credentials, expanded secret scanning and network restrictions, redeployed production infrastructure, rebuilt the extension marketplace, and brought in external forensic investigators.
What Fireworks says — and what is still open
One attribution gap matters. The chain from the exposed key to the tampered extension is Cursor's account. Fireworks has published two bulletins about its own October incident, on 9 October and 9 November, and neither names Cursor. Fireworks disclosed two intrusions — one detected on 6 October, and a second on 23–24 October through a third-party cloud provider. It is not public which one exposed Cursor's key.
Fireworks' 9 November update says it later found impact to account and billing information and, for a limited set of accounts, to secrets, datasets and incomplete model weights. Payment card numbers are not stored and were not impacted. Fireworks says it notified law enforcement and brought in Palo Alto Networks Unit 42 for forensics.
What to do if you use Cursor
If you updated the Claude Code extension through Cursor on 29 October, check for version 2.1.312 or 2.1.313, and reinstall from a trusted source if you did. If you use Cursor without Privacy Mode, treat any API keys in your prompts as potentially exposed and rotate them.
Key takeaways
- Cursor says a key it issued to Fireworks AI was exposed at Fireworks, and an attacker used it to copy a bucket of coding-agent traces, mostly from 2025.
- Some traces held unredacted credentials — under 0.01 percent of customers had API credentials exposed; traces from under 0.05 percent of users were copied.
- On 29 October the attacker placed malicious copies of the Claude Code extension (versions 2.1.312 and 2.1.313) on Cursor's marketplace cache for about 11 hours; up to 5,000 users may have received them.
- Cursor says Privacy Mode data, code repositories, production systems and the Cursor app were not affected, and the last attacker activity was on 1 November.
- The chain is Cursor's account — Fireworks' own bulletins do not mention Cursor, and it is not public which of Fireworks' two intrusions exposed the key.
Conclusion
This was not a breach of Cursor's front door. It was a breach of a side door — a vendor key, a shared bucket, a marketplace cache — and it reached users' machines anyway. The pattern is the warning: in AI coding tools, your prompts and your secrets travel further than you think. Rotate exposed keys, check that 29 October update, and turn on Privacy Mode.
FAQ
Was I affected by the Cursor breach?
Most were not. Traces from under 0.05 percent of users and API credentials from under 0.01 percent of customers were exposed. If you updated the Claude Code extension on 29 October, check for version 2.1.312 or 2.1.313.
What is the Fireworks AI connection?
Fireworks AI was a Cursor inference vendor. Cursor says a key it issued to Fireworks was exposed there, and an attacker used it to copy a shared bucket of coding-agent traces. Fireworks' bulletins do not mention Cursor.
Is the Cursor incident contained?
Cursor says yes: the key is revoked, inference has moved off Fireworks, credentials are rotated, the marketplace is rebuilt, and the last attacker activity it saw was on 1 November.
Researched with AI assistance and reviewed for accuracy.
Sources
- Cursor — An update on our recent security incident, 13 November 2026
- iFynx — Cursor Security Breach: Fireworks Key and Fake Claude Code Extension
- Abijita — Cursor Details Breach Behind Malicious Claude Code Extension
- Daily Tech & AI — Claude Code Leak: Check Your GitHub for Secrets Now
- Daily Tech & AI — Microsoft Lazarus: 7 Neurons Can Backdoor AI
- Daily Tech & AI — GPT-6 Cyber: OpenAI's New Security AI Explained
Follow on WhatsApp
Follow Daily Tech & AI on WhatsApp for one tech and AI update every day: Daily Tech & AI WhatsApp channel.
Comments
Post a Comment