Fireworks AI reports two security incidents as Cursor details exposed key
Fireworks AI said the activity involved two separate security incidents, spanning August 3 to October 7 and October 23 to 30, 2026. Cursor separately said a Fireworks-exposed key was used to copy data from a storage bucket.

Key takeaways · 4
- 01
Fireworks described two incidents with different time windows and access paths; assess them separately when reviewing exposure.
- 02
The second incident involved account and billing information for accounts created before October 30, but not payment-card numbers.
- 03
Cursor said the Fireworks-exposed key was used to copy data from a read-only bucket, and that it disabled the key.
- 04
Cursor estimated that at most 5,000 users requesting specific cached extension versions on October 29 potentially received a malicious build.
Fireworks described two incidents
Fireworks said the activity involved two separate security incidents.[1] The first ran from August 3 to October 7, 2026, and involved stolen credentials and a service vulnerability being used to misuse the platform.[1] Fireworks said secrets, model files and datasets belonging to a small number of customers were exposed.[1] It said Unit 42 completed its forensic investigation of that incident and described it as contained and remediated.[1] The second incident ran from October 23 to October 30.[1] Fireworks said it began with unauthorized access to development machines hosted by a third-party cloud provider that had experienced a security incident.[1]
Second incident included account data
Fireworks said the intruder used access to the development machines to reach part of its production systems.[1] Account and billing information for all accounts created before October 30 was accessed, along with datasets, models and/or secrets belonging to a small number of customers.[1] Fireworks said the billing information did not include payment-card numbers because it does not store those numbers.[1] AprosSyber was conducting a forensic investigation of the second incident, which Fireworks described as contained and remediated.[1]
Cursor detailed the exposed key
Cursor said a key exposed by Fireworks was used by an attacker to copy data from a read-only storage bucket.[2] The bucket contained limited coding-agent traces, mostly from 2025 sessions that were not in Privacy Mode and were permitted for model training.[2] Cursor said some traces contained unredacted credentials, and API credentials were exposed for less than 0.01% of its customers.[2] It also listed accessed data including archived non-production source repositories, support data, email conversations, usage metrics, contact details, conversation summaries and credentials sent to support.[2] Cursor said traces were exposed for fewer than 0.05% of users not using Privacy Mode and that it identified and directly notified those users.[2]
Marketplace incident and response
Cursor said malicious copies of Anthropic’s Claude Code extension were placed in the cache serving Cursor’s VS Code extension marketplace on October 29.[2] The malicious build was available for approximately 11 hours, and Cursor estimated that at most 5,000 users requesting specific cached versions that day potentially received it.[2] Cursor said Anthropic did not publish the malicious copy and that the Open VSX marketplace source never contained the malware.[2] Cursor said it disabled the Fireworks key, moved inference traffic off Fireworks and cut off Fireworks’ access to shared datasets.[2] It said the last attacker activity it had seen was on November 1 and was blocked.[2]
For teams assessing vendor exposure, the disclosures describe more than one incident and different categories of affected data, so treating them as a single event could obscure what needs review. Cursor’s account also gives customers concrete exposure measures and response steps to consider when checking their own records and vendor dependencies.
Why it matters
AI training for your team
kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.
How this developed
21 November 2026
Fireworks AI reports two security incidents as Cursor details exposed key
Sources
- Security bulletinsfireworks.ai
- An update on our recent security incident · Cursorcursor.com