Fireworks AI Discloses Wider Breach Reaching Datasets and Model Weights for Some Customers
Fireworks AI said on Nov. 9 that an October breach reached customer billing data and, for some accounts, datasets, model weights and secrets.

Fireworks AI, an AI infrastructure company based in San Mateo, said on Nov. 9, 2026 that an October security incident had reached more customer data than it first reported. Its updated notice says attackers accessed billing information and, for a limited set of accounts, secrets, datasets and incomplete model weights. The Record and BleepingComputer reported the update on Nov. 10.
The disclosure covers two intrusions in October. The second entered through a cloud provider that hosted Fireworks' research and development machines, according to the company's bulletin, and as of Oct. 25 the company had not detected further evidence of successful unauthorized access to the platform its customers use.
Two October Intrusions With Different Entry Points
The first intrusion began with unauthorized use of internal Fireworks credentials, according to The Hacker News. It gave the attacker access to environment variables in evaluation jobs for a limited number of customer accounts. The Record gives Oct. 6 as the date the breach was first discovered. The date of Fireworks' first bulletin differs by source: the company's update page lists Oct. 9, while The Hacker News gives Oct. 10.
The second intrusion began on Oct. 23, when a third-party cloud provider hosting Fireworks' R&D machines had a security incident, according to the company's bulletin. On Oct. 24, the attacker used that access to reach Fireworks development machines and run an unauthorized workload that caused additional customer data impact. Fireworks says it detected that activity the same day. As of Oct. 25, it had seen no further evidence of successful access to the platform customers use. The Hacker News reports that the cloud provider is not named.
Which Data Was Accessed and Who Is Affected
Fireworks' bulletin lists three categories of exposed data: resource metadata such as identifiers, display names, descriptions and creation times; billing information including addresses, invoice and payment history, account balance and credit usage; and, for certain customers, a subset of incomplete model weights and datasets. The bulletin also says secrets were affected for a limited set of accounts. The company says it does not store payment card numbers, and they were not impacted.
Fireworks has not given a count of affected customers. BleepingComputer describes the group as "some customers," while The Record says the company described a limited number of customer accounts. The Record says impacted customers have been notified, and Fireworks' bulletin says it is notifying customers whose data was impacted.
Notion, a named customer, said fewer than 1% of its workspaces were affected, according to The Hacker News. The Record reports that Notion was notified on Nov. 2 and disconnected from Fireworks. The Hacker News adds that Notion AI chat transcripts used for its Insights feature were included. Notion has disabled that feature and stopped serving open-weight models through Fireworks.
How the Company and Investigators Describe the Attacker
Fireworks' bulletin links the activity to a financially motivated cybercriminal group that sells access to LLM tokens obtained through unauthorized access to illicit routing gateways. The Record reports that investigators found a financially motivated cybercriminal group launched the attack, and Fireworks had hired Palo Alto Networks Unit 42 to investigate. The group is not named in the reports.
The Hacker News assesses the activity as part of LLMjacking, a scheme in which stolen credentials are used to run up costs on a victim's LLM services while criminals resell that access. The company's bulletin, as reported by The Hacker News, says it revoked compromised internal credentials, rotated service account tokens and revoked affected API keys, including keys the attacker created. The bulletin also says Fireworks notified law enforcement, isolated affected systems and tightened access to data storage and analytics.
“The second intrusion entered through a third-party cloud provider; as of Oct. 25, Fireworks had not detected further access to the platform customers use.”
Service Limits and Open Questions
Fireworks says self-serve dataset uploads and downloads for Batch Inference and dataset-based training are disabled while the investigation runs. The Training SDK remains available under zero data retention, and Lab-based training runs in an isolated environment. The company says it has not detected any impact to Inference, the Training SDK or Nexus, which are zero data retention by default.
As of the Nov. 9 update, the investigation was ongoing, and Fireworks said it would provide further updates as appropriate. The reports do not give a date for completion or a final count of affected accounts.



