The chain, in order
Cursor's account is a sequence, and the order is the point. A key Cursor issued to Fireworks was exposed by Fireworks. That key opened a read-only bucket Cursor used to share data with the vendor. The bucket contained coding-agent traces, mostly from 2025 sessions outside Privacy Mode, which Cursor says it was permitted to use for model training. A small number of those traces contained credentials that had not been redacted. Those credentials opened select SaaS tools Cursor uses. From there the attacker reached archived non-production source code repositories, limited customer support data and email conversations, usage metrics, summaries of some coding-agent conversations, and some credentials users had sent to Cursor's support team.
Only then came the part users saw: on October 29, a stolen credential was used to place malicious, unauthorized copies of Anthropic's Claude Code extension directly on the cache serving Cursor's extension marketplace. Anthropic did not publish those copies, Cursor stresses, and the upstream marketplace, Open VSX, never carried them.
Read the percentages both ways
Cursor reports trace exposure for fewer than 0.05% of users and API-credential exposure for fewer than 0.01% of customers; those figures do not describe every impact of the incident, which also reached internal tools and a software distribution channel. Cursor says it contacted each affected customer directly, emailing team admins the specific keys to rotate where credentials were involved.
Read the same numbers from the other end. At most 5,000 users potentially received a malicious build of the Claude Code extension, an upper bound on delivery rather than a count of confirmed installs, and Cursor's own notice says uninstalling it may not be enough: anyone who installed or updated to versions 2.1.312 or 2.1.313 from Cursor's marketplace during the window is asked to follow the remediation steps in its October 30 security notice. A percentage describes the denominator. The 5,000 describes the exposure the remediation has to cover.
The vendor is the finding
The new information in this report is not the extension. That was disclosed on October 30, when ALFW covered the initial notice. The finding is that the intrusion began at a vendor, through a key the customer issued and data the customer chose to share, including traces with unredacted credentials inside them.
Cursor's response reflects that. It did not only rotate the exposed key. It moved all inference traffic off Fireworks, cut off Fireworks' access to every shared dataset, rotated credentials across its internal systems even where it had no evidence of exposure, added secret scanning across repositories and other internal data sources, and redeployed its production infrastructure. Fireworks is also the vendor at the center of two other disclosures ALFW has covered this month: its own expanded breach finding on November 9 and Notion's confirmation that AI chat transcripts were accessed through it. This report supplies what neither of those did: a detailed causal account from the customer side, ending with the vendor's inference role and data access terminated.
What remains open
Cursor says it believes the incident is contained, with the last attacker activity seen on November 1 and blocked, and that monitoring continues. Two things are not in the post. It does not say how the key came to be exposed at Fireworks, which sits inside Fireworks' own incident, disclosed separately. And it does not quantify how many of the at most 5,000 potentially served users actually installed a malicious build, only that affected users and team admins were notified starting October 30.
If your team used Cursor in late October, the practical checks are specific: look for a notification from Cursor, and check whether anyone installed or updated the Claude Code extension to 2.1.312 or 2.1.313 from Cursor's marketplace on October 29. Extension installation and trace exposure are separate questions; consult Cursor's notifications and the applicable remediation notice.
Sources
- [1] Cursor, "An update on our recent security incident" (November 13, 2026)Read source
- [2] Cursor, security notice and remediation steps (October 30, 2026)Read source
- [3] Unite.AI, "Cursor Traces Breach to Access Key Exposed by Fireworks AI"Read source