Raw data, clear context.

[
[
[

]
]
]

Cursor’s 13 November investigation update traces its October extension incident to an access key exposed at Fireworks AI.[1] The reconstruction adds the credential trail behind the tampered extension covered in our earlier report.[1]

According to Cursor, the key allowed copying from a read-only shared bucket holding coding-agent traces, mostly from 2025 and sessions outside Privacy Mode, retained for model training.[1] Unredacted secrets in some traces opened SaaS tools, archived non-production repositories, support material and further credentials; a stolen internal credential then enabled marketplace-cache tampering on 29 October.[1]

Fireworks’ updated bulletin expands its earlier findings to account and billing data, plus datasets and incomplete models for certain customers.[4] It reports development-machine access on 24 October following a cloud-provider incident; its investigation remains ongoing.[4]

Cursor reports trace exposure for fewer than 0.05% of users and API-credential exposure for fewer than 0.01% of customers.[1] At most 5,000 users potentially received selected cached builds on 29 October 2026.[1] Its update gives no count of confirmed endpoint compromises. SecurityWeek reports Cursor’s account without providing an independent forensic reconstruction.[2]

Cursor-reported route from a Fireworks-exposed key through historical traces to internal credentials and cache tampering. Separate figures: fewer than 0.05% of users had traces exposed, fewer than 0.01% of customers had API credentials exposed, and at most 5,000 users potentially received selected cached builds on 29 October 2026. Cursor gives no count of confirmed endpoint compromises.
Cursor’s reported route and separate exposure measures. The 5,000-user ceiling concerns potential receipt of selected cached builds on 29 October.[1]

Cursor says Privacy Mode prompts and code, active inference requests and customer code repositories were unaffected.[1] It has withdrawn Fireworks’ access, moved all inference traffic away from the supplier and cut off access to shared datasets.[1] It believes the incident is contained; the last attacker activity it observed, on 1 November, was blocked.[1]

The company says it emailed administrators specific exposed keys to rotate.[1] For anyone who installed or updated to Anthropic.claude-code 2.1.312 or 2.1.313 from Cursor’s VS Code marketplace on macOS or Windows between 02:27 and 13:35 PT on 29 October 2026, its security notice warns that uninstalling alone may leave persistence behind.[3] It directs affected users to check their systems, remove persistence and rotate credentials accessible from affected hosts from a trusted system.[3]

Sources

  1. An update on our recent security incident · Cursor
  2. Cursor Supply Chain Attack Started at Fireworks AI – SecurityWeek
  3. Security notice — malicious version of Anthropic.claude-code (2.1.312 and 2.1.313) identified in the Cursor VS Code Extension Marketplace · Cursor
  4. Update on October 2026 security incident | Fireworks AI