Fireworks AI Data Breach Investigation

A Notion customer notice provided to Hall Attorneys says the recipient’s workspace was among fewer than 1% of Notion workspaces that may have been affected by the Fireworks incident. It describes possible exposure of some Notion AI chat sessions. This is an attributed customer notification, not a finding that every user or chat in a notified workspace was involved.

The notice says the chats lacked author and workspace labels and did not include tool-call results. It says no other data within the workspace was affected. These statements concern the notice’s described scope; they are not our independent verification of the full incident or any particular person’s exposure.

The notice’s term unattributable refers to missing labels. It does not establish that the words in a transcript contain no names, personal details, credentials, or confidential business information. A workspace percentage is not a count or percentage of affected people.

Notion’s official subprocessor list identifies Fireworks as an AI infrastructure provider. That relationship alone does not establish exposure. A public recipient post reproduces similar notification language, but it is not an official public incident advisory. The excerpt provided for this review does not establish a complete chat-exposure window.

The updated notice identifies resource metadata and billing records, including addresses, invoice and payment history, balances, and credit usage. It also identifies access to some customers’ datasets and incomplete model weights. These categories do not establish that every account or every record within an account was accessed.

Fireworks says payment-card numbers were not stored or impacted and reports no detected impact to Inference, Training SDK, or Nexus. Those are product-specific findings from an ongoing investigation, not a guarantee that every Fireworks service, stored file, prompt, or customer was unaffected.

The November update expands the earlier assessment. The original bulletin’s statements about no detected model, dataset, or personal-information access should not be treated as current blanket exclusions. Default zero-data-retention settings do not establish protection for every product, workflow, or stored dataset.

We have not independently determined the affected-person total, each account’s exposure, whether a particular credential was misused, or the extent of any downstream access. A customer relationship, an unfamiliar charge, or use of an app connected to Fireworks does not establish that connection by itself.

Fireworks’ product-specific findings and the Notion notice describe their respective reviews. We have not mapped the reported Notion chat material to a particular Fireworks storage or processing path. A general product finding should not be used to disregard a workspace-specific notice, and that notice does not establish exposure of all Notion content.

Notion users should verify a notice through a known workspace administrator or official Notion support. If your account offers a review of potentially affected material, use verified account instructions rather than an unsolicited link. If a relevant chat contained credentials, work with your security team and the issuing provider to rotate them; do not send those values to us.

The November update says self-service dataset uploads and downloads for Batch Inference and dataset-based training were disabled during the investigation. Check current official instructions for your product rather than treating a dated notice as a guarantee of service availability.

The original bulletin directs affected customers to replace revoked Fireworks keys needed by their applications and rotate identified third-party credentials with the providers that issued them. Removing a stored secret does not revoke the underlying credential. Follow current account-specific instructions and update dependent applications.

Review associated access logs and resources you are authorized to inspect, including unfamiliar keys, users, evaluation or training activity, and datasets. A key may require rotation even when no suspicious activity is visible. Work with your organization’s security team and the issuing provider as appropriate.

Use the official bulletin or a known provider contact to verify notices. The November update gives security-notice@fireworks.ai for incident questions. Requesting a legal review does not submit a support ticket, revoke credentials, restore data, or stop unauthorized use; do not delay protective action while awaiting a review.

Keep notices, relevant invoices, account alerts, suspicious messages, support correspondence, and records of credential changes, restrictions, and response costs. For Notion, preserve the original notification and its date and note your workspace role and whether you used Notion AI. Note whether you were a direct customer, an administrator, a business representative, or a user of another application. Describe logs and account records before sending copies.

Do not send API keys, tokens, passwords, secret values, environment-variable files, model weights, training or evaluation datasets, confidential prompts or outputs, customer records, or complete payment information in an initial inquiry. Keep only records you are authorized to access. We can discuss what is needed and an appropriate way to provide it.