🔍 Read the full analysis: How AI Customers Can Review Their Risk From The Fireworks Breach on ThorstenMeyerAI.com
Get the latest gadgets 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 November 9 update says an October security incident affected account and billing information and, for a limited set of accounts, secrets, datasets and incomplete models. Customers should check what data and credentials were connected to Fireworks; the company has not said how many customers were affected or named the third-party cloud provider involved.
Fireworks AI said on November 9 that an October security incident reached further than initially reported, exposing account and billing information and, for a limited set of accounts, secrets, datasets and incomplete model weights. The update matters to companies using Fireworks to run or fine-tune AI models because their credentials and proprietary data may have been available through affected systems.
Fireworks’ first public bulletin, published October 9, described the misuse of internal credentials to read environment variables from a limited number of customer evaluation jobs. Those variables can contain API keys and other credentials. At the time, the company said it had not detected unauthorized access to customer models, datasets, training data, inference traffic or personal information. Its November 9 update revised that picture: it said the affected internal application provided access to resource metadata and billing details, and that some customer accounts also had secrets, datasets and incomplete models affected.
Fireworks described two stages of activity. On October 6, it identified misuse of internal credentials and began containment. It said an attacker later used a third-party cloud provider’s security incident to reach Fireworks development machines and run an unauthorized workload on October 24. Fireworks detected that activity the same day. The company said it had not detected further successful unauthorized access on the platform customers use after October 25. It also said it notified law enforcement and engaged Palo Alto Networks Unit 42 to assist with the investigation.
The company says exposed billing information could include billing and shipping addresses, invoice and payment history, account balances and credit usage. For some customers, a subset of incomplete model weights and datasets was accessed. Fireworks says it does not store payment card numbers and that it has detected no impact to its Inference, Training SDK or Nexus products, which it describes as zero-data-retention by default. As a precaution, it disabled self-service dataset uploads and downloads used by Batch Inference and dataset-based training while investigating.
Notion separately said chat transcripts from its Notion AI Insights feature were included in a dataset attackers may have accessed. Notion said fewer than 1% of its workspaces were affected. That statement concerns Notion’s own assessment; Fireworks has not disclosed an overall customer count or a complete list of affected accounts.
Review your Fireworks risk after the breach
Fireworks AI’s November 9 update expanded what may have been exposed in an October incident. The impact varied by account; customers should check which credentials and data were connected to affected workflows.
What may have been exposed
The reported impact varied by account and included several types of business information.
Billing and resource details
Billing information may include addresses, invoices, payment history, balances and credit usage.
Secrets in some accounts
Job environment variables can contain API keys and other credentials.
Datasets and incomplete models
Fireworks said a limited set of accounts had datasets and incomplete model weights affected.
How the disclosure changed
What customers should review
Match the reported categories to your own Fireworks use and ask for account-specific details.
Review evaluation jobs, dataset-based training, Batch Inference, resource descriptions and billing records.
Ask Fireworks whether your account was affected and which specific resources or files were involved.
Check whether job secrets should be replaced and whether they could grant access to other systems.
What remains unclear
- No further successful access on its customer platform after October 25
- No detected impact to Inference, Training SDK or Nexus
- Payment card numbers are not stored by Fireworks
- Overall customer and account count
- Name of the third-party cloud provider
- Meaning and scope of “incomplete models”
- Whether accessed files were copied or misused
What AI Customers Should Check
The incident shows why customers of AI infrastructure providers need to assess not only whether prompts were retained, but also what credentials, evaluation data and training materials were available to connected systems. Even where a service is designed for zero data retention, other features or internal applications may handle different categories of information. Fireworks’ account does not establish that all customer prompts or model files were accessed; it says the impact varied by account and involved a limited set for secrets, datasets and incomplete models.
Organizations that used Fireworks should compare the company’s affected categories with their own use: evaluation jobs, dataset-based training, batch inference, stored resource descriptions and billing records. They can ask Fireworks whether their account was affected, which specific resources or files were involved, and whether credentials in job environments should be replaced. Companies that used secrets with other systems may also need to assess whether those credentials granted access beyond Fireworks. These are prudent review steps, not evidence that any particular customer suffered misuse.
The distinction between incomplete weights and finished models also matters. Model weights can embody the results of training on a company’s data, but Fireworks has not explained what “incomplete” means in this incident. The available information therefore does not show whether accessed files could be used to reproduce a customer’s model or reveal its training data.
Products Worth Considering

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight

Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black
As an affiliate, we earn on qualifying purchases.
How the Disclosure Changed
Fireworks’ October 9 bulletin said it had identified activity on October 6 and started notifying affected customers on October 7. The company reported rotating employee API keys on October 9 and revoking additional user API keys and rotating service-account tokens on October 10. Its initial assessment said it had not detected access to customer datasets or models. The November 9 update added the later cloud-provider route and expanded the reported data categories.
According to Fireworks, the third-party provider experienced a security incident on October 23, and an unauthorized party used that access to reach Fireworks development machines the following day. Fireworks says it detected the workload on October 24 and saw no further successful access by the threat actor on its customer platform after October 25. The company’s revised notice came more than a month after the first detection, following further investigation.
Notion’s account adds a customer-specific example of the review process. It said Fireworks first contacted it on November 2 about a dataset thought to contain synthetic test data, then warned on November 6 that potential customer data may have been accessed. Notion said the relevant dataset was confirmed on November 8. Its statement identifies one potential data set and a share of workspaces, but does not establish the impact for other Fireworks customers.
“For a limited set of accounts, we also discovered impact to secrets, datasets, and incomplete models.”
— Fireworks AI, November 9 security update
Open Questions on Exposure
Several central details remain undisclosed. Fireworks has not named the third-party cloud provider, said how many customers or accounts were affected, or explained how the internal credentials used in the earlier activity were first obtained. It has not described what happened to accessed data, whether files were copied outside the affected systems, or whether anyone has tried to use the information.
The scope of “incomplete models” is also not defined. Fireworks has not said whether the phrase refers to unfinished training runs, partial model files or another form of data, nor whether the datasets and secrets were accessed in the same accounts. The company says its investigation is ongoing and may reveal additional information. Its statement that it detected no impact to certain products and no further successful access after October 25 is a report of its current findings, not a final account of every possible consequence.
Customer Reviews and Investigation
Fireworks says it is continuing its investigation with assistance from Unit 42 and has temporarily disabled certain self-service dataset uploads and downloads. The company has not announced when those functions will return or given a date for a final report. Customers can seek account-specific information from Fireworks and review whether credentials, datasets or model materials were present in the affected workflows.
Companies should also watch for further notices from Fireworks and any service providers or product teams that relied on its infrastructure. The next meaningful developments will be a clearer account of affected customers and data, any explanation of the cloud-provider incident, and updates on the investigation’s findings. Until those details are available, the public record supports targeted customer reviews, but not a conclusion that every Fireworks user’s data was exposed.
Key Questions
What information does Fireworks say was affected?
Fireworks says the incident involved account and billing information, including addresses, invoices, account balances and credit usage. For a limited set of accounts, it also reported effects on secrets, datasets and incomplete model weights.
Were payment card numbers exposed?
Fireworks says it does not store payment card numbers and that they were not impacted. This is the company’s statement in its November 9 update.
Does the update mean all Fireworks customers’ prompts were accessed?
No. Fireworks has not said that all customers’ prompts or inference traffic were accessed. It said it detected no impact to its Inference, Training SDK and Nexus products, and described datasets and secrets as affected for a limited set of accounts.
What should a Fireworks customer review?
Customers can ask Fireworks whether their account or resources were affected and review credentials, evaluation-job environment variables, datasets, model files and billing records connected to the platform. They should treat this as a precautionary review, not proof that their information was accessed.
What is still unknown about the breach?
Fireworks has not named the third-party cloud provider, given a total number of affected customers, explained how the internal credentials were obtained or said what happened to the accessed data. It also has not defined what it means by incomplete models.
Source: ThorstenMeyerAI.com
