🔍 Read the full analysis: Fireworks AI Breach: Are Your Datasets Or Model Weights Involved? on ThorstenMeyerAI.com
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Fireworks AI’s Nov. 9 update says its security incident reached account and billing information and, for a limited set of accounts, secrets, datasets and incomplete models. Notion says chat transcripts from its AI Insights feature may have been in an affected dataset, involving fewer than 1% of its workspaces. Fireworks says it has not detected impact to several named products, but the number of affected customers and the fate of accessed data remain unknown.
Fireworks AI said Nov. 9 that an incident first disclosed in October reached account and billing information and, for a limited set of accounts, secrets, datasets and incomplete models. The company’s expanded account matters to businesses using its AI services because customer data and model assets may have been accessed; the number of affected customers is not yet public.
In its update, Fireworks said the incident involved two routes into its systems. On Oct. 6, it identified activity involving internal Fireworks credentials used to read environment variables from customer evaluation jobs. Environment variables can contain settings and credentials, such as API keys, though the company’s update does not specify which particular secrets were accessed or how the credentials were first obtained.
A separate event involved a third-party cloud provider that hosted Fireworks research and development machines. Fireworks said the provider experienced a security incident on Oct. 23, and that an unauthorized party used access to reach Fireworks development machines and run an unauthorized workload on Oct. 24. The company said it detected that activity the same day. It reported no further evidence of successful unauthorized access on its customer platform after Oct. 25.
Fireworks lists resource metadata, billing information and, for certain customers, a subset of incomplete model weights and datasets among the affected information. It says payment card numbers were not affected because it does not store them. It also says it has detected no impact to Inference, Training SDK or Nexus, which it describes as zero-data-retention by default. Fireworks said it disabled self-serve dataset uploads and downloads used by Batch Inference and dataset-based training while it investigates.
Fireworks AI breach What may be involved?
Fireworks says a limited set of accounts may have had secrets, datasets and incomplete model weights accessed. The customer count and what happened to accessed data remain unknown.
What may be involved
The update identifies several kinds of information and assets, with limited account scope for some.
Account and billing information
Fireworks reports impact to resource metadata and billing information.
Secrets, datasets and model weights
A subset of datasets and incomplete model weights may have been accessed. Fireworks has not defined “incomplete.”
AI Insights transcripts
Notion said chat transcripts may have been in an affected dataset; its workspace estimate does not represent all Fireworks customers.
Two routes into systems
Fireworks described separate activity involving internal credentials and a third-party cloud provider.
On Oct. 6, Fireworks identified activity using internal credentials to read environment variables from customer evaluation jobs.
After a cloud provider incident on Oct. 23, an unauthorized workload ran on Fireworks development machines on Oct. 24.
Service impact reported
Fireworks says it detected no impact to these named products.
| Product or information | Fireworks’ update |
|---|---|
| Inference | ✓ No impact detected |
| Training SDK | ✓ No impact detected |
| Nexus | ✓ No impact detected |
| Payment card numbers | ✓ Not stored; not affected |
What remains unknown
- How many Fireworks customers were affected, and which accounts were involved.
- Whether accessed data or model artifacts were copied, retained, shared or used.
- What “incomplete” means for the affected model weights.
- Which specific secrets, datasets and customer assets were involved.
Risks to Customer Data and Models
The disclosure is relevant to companies that use hosted AI services to evaluate, train or run models. Such services can handle business datasets, credentials and model artifacts; exposure can create risks beyond ordinary account information, including potential access to material a customer uses to build or operate an AI product. The update does not establish that any accessed material was used, published or copied for later misuse.
Notion has provided a specific example of potential customer data in scope. It said chat transcripts from its Notion AI Insights feature were in a dataset attackers may have accessed, and that fewer than 1% of its workspaces were affected. That figure refers to Notion workspaces, not the total number of Fireworks customers or all organizations affected by the incident.
Fireworks’ statement that certain model weights were incomplete leaves customers with an important unanswered question: what portion of each model artifact was involved, and whether it could be useful outside the training process. The company has not defined “incomplete” in the update. Until that detail and the affected customer list are clearer, organizations may need to check notices from Fireworks and assess whether their own data or assets were involved.
Products Worth Considering
Bose Lifestyle Ultra Compact Wireless Smart Home Speaker Bundle, Black
Dyson Omni-Glide Cordless Stick Vacuum, Gold/Iron
Arlo Pro Wireless Security Camera, 2K HDR, 6th Generation, 4-Camera System
As an affiliate, we earn on qualifying purchases.
From October Disclosure to Update
Fireworks’ Oct. 9 bulletin described the first part of the incident. At that point, it said it had not detected unauthorized access to customer models, datasets, training data, inference traffic or personal information. Its Nov. 9 update revised that picture, saying it later discovered impacts to account and billing information and limited access to secrets, datasets and incomplete models.
Fireworks’ timeline says it began containment on Oct. 6 and started notifying affected customers on Oct. 7. It rotated employee API keys on Oct. 9, then revoked more user API keys and rotated service-account tokens on Oct. 10. The second intrusion followed the cloud provider incident later in October. Fireworks said it notified law enforcement and engaged Palo Alto Networks Unit 42 to conduct forensic work.
Notion said Fireworks first contacted it on Nov. 2 about a small dataset believed to be synthetic test data. Fireworks later told Notion, on Nov. 6, that malicious actors may have accessed potential customer data; Notion said the dataset was confirmed on Nov. 8. Fireworks published its broader update the next day.
“For a limited set of accounts, we also discovered impact to secrets, datasets, and incomplete models.”
— Fireworks AI, in its Nov. 9 security update
Scope and Data Use Still Unknown
Fireworks has not identified the third-party cloud provider, said how many customers were affected or explained how attackers first obtained the internal credentials involved in the October activity. It has also not reported whether data or model artifacts were exfiltrated, retained, shared or used after access. Access does not by itself establish that information was taken or misused.
The company has not clarified what “incomplete” means for the model weights, or specified the exact datasets, secrets or customer accounts involved. Notion’s disclosure identifies one possible dataset and gives a workspace estimate, but does not resolve the broader scope of the Fireworks incident. The investigation is ongoing, and Fireworks says further information may emerge.
Investigation and Customer Notices
Fireworks says it is continuing its investigation with assistance from Unit 42 and has notified law enforcement. It has temporarily disabled certain self-serve dataset uploads and downloads while that work proceeds. Customers should look for direct notices from Fireworks and review any guidance the company provides about affected accounts, credentials, datasets or model artifacts.
The next meaningful update would clarify the number and identity of affected customers, the categories and extent of data accessed, and whether investigators found evidence of copying or subsequent use. Fireworks has not given a date for a final report. Its present account should be treated as a developing description rather than a complete inventory of affected information.
Key Questions
Were customer datasets or model weights involved?
Fireworks says yes for a limited set of accounts: a subset of datasets and incomplete model weights was accessed. It has not named the affected customers or defined how much of each model was involved.
Were payment card numbers exposed?
Fireworks says it does not store payment card numbers and that they were not affected. It separately reports impact to billing information, including billing and shipping addresses, invoice and payment history, account balances and credit usage.
Was Notion data involved?
Notion said chat transcripts from its AI Insights feature were in a dataset attackers may have accessed. Notion reported that fewer than 1% of its workspaces were affected; this is not an estimate of all Fireworks customers.
Which Fireworks services does the company say were not affected?
Fireworks says it has detected no impact to Inference, Training SDK or Nexus, which it describes as zero-data-retention by default. That statement reflects the company’s findings so far, while its investigation remains ongoing.
What should Fireworks customers do now?
Customers should check for direct notices from Fireworks and follow any account-specific instructions. The company has not published a complete affected-customer list or a final account of what was accessed.
Source: ThorstenMeyerAI.com
