Fireworks AI October security incidents

October activity exposed secrets and other customer data, with downstream impact to Notion AI Insights.

Last modified

Summary

  • Environment: Evaluation secrets, development machines and stored datasets
  • Operational impact: Dataset transfers and Notion Insights disabled during investigation
  • Financial impact: No confirmed incident cost quantified
  • Record status: Developing record. Reviewed November 10, 2026; updated as evidence emerges.

What happened

Fireworks AI detected misuse of internal credentials on October 6, exposing evaluation-job secrets in a limited set of customer accounts. Its initial bulletin was published October 9. [3]

A November 9 update described further activity: attackers reached development machines through an unnamed cloud provider on October 24 and ran an unauthorized workload. [2]

Impact

  • Fireworks broadened its findings to resource metadata, billing information, and—for some accounts—secrets, datasets and incomplete model weights. Card numbers were excluded. The update does not assign every category to a particular intrusion. [2]
  • Notion reported affected Insights chat transcripts in fewer than 1% of workspaces. Most lacked user/workspace labels and tool-call results; the text itself could still contain identifying or sensitive information. [1]

Timeline

  1. Credential misuse detected

    Containment began the same day.

    [3]
  2. Customer notification begins

    Affected customers began receiving notices.

    [3]
  3. Development-machine activity detected

    Fireworks detected the unauthorized workload following the provider’s October 23 compromise.

    [2]
  4. Notion disconnects Fireworks

    The initial November 2 warning concerned synthetic testing data.

    [1]
  5. Additional dataset confirmed

    Following a November 6 warning, Notion identified customer-chat impact in the confirmed dataset.

    [1]
  6. Briefing updated

    This briefing was last reviewed and updated on November 10, 2026.

Threat Group & Attack Vector

The first activity involved internal credentials; the later entry used a compromised hosting provider. Fireworks has not identified that provider or a specific exploited vulnerability. [2][3]

Actors

  • No threat actor group has been identified in the reviewed public evidence.

TTPs

Response

Fireworks revoked keys and requested rotation of exposed third-party credentials. [3] It engaged Unit 42, restricted systems and disabled self-service dataset transfers. Its investigation remained open. [2]

Notion’s advisory says Insights was disabled, Fireworks-based open-weight inference stopped, and affected customers contacted with access to their impacted data. [1]