Security
Fireworks AI says a second October intrusion via a third-party cloud provider reached billing records and, for some accounts, secrets, datasets and incomplete model weights: no card numbers were affected, and Forbes reports a criminal group
Fireworks AI says an October intrusion reached billing data and, for some accounts, datasets and incomplete model weights. No cards; scope unclear.
Fireworks AI, the Nvidia-backed inference platform, updated its security notice on November 9 to say an October intrusion went beyond what it first reported. Attackers reached billing records and, for a limited set of accounts, stored secrets, datasets and incomplete model weights. The company says payment card numbers were not stored or affected, and has not said how many customers are involved.
The essentials
- What changed: on October 6 Fireworks found internal credentials had been used to read environment variables in some customer accounts, and on October 10 it said it had found no evidence that models or datasets were accessed. The November 9 update contradicts that for the later incident, per The Neuron.
- The second intrusion: after a security incident on October 23 at a third-party cloud provider that hosted its research machines, an attacker accessed those machines on October 24 and ran an unauthorized workload, per GBHackers. Fireworks says it detected this the same day and has seen no further successful access to the customer platform since October 25.
- What was exposed: resource metadata and billing information (addresses, invoices, payment history, balances, credit usage) for affected accounts; for certain customers a subset of incomplete model weights and datasets; for a limited set of accounts, secrets. No card numbers, and no claim that complete models were stolen.
- Who and why: Technology.org relays that reporting links the attack to a financially motivated group that steals and resells access to AI models, known as LLMjacking, and The Neuron says Forbes reported two customers moved away. Fireworks notified law enforcement and hired Palo Alto Networks Unit 42 for forensics, per GBHackers.
Why the wording matters
“Incomplete” weights and “a limited set” of accounts limit what can be inferred. The number of affected customers, the volume of data and whether any of it was used are not published, and customers should treat stored secrets as exposed.
What customers can do
Rotate any API keys and secrets stored on the platform, review users, jobs, datasets and keys for unfamiliar activity, and watch for phishing that uses billing details. This is general guidance relayed from coverage, not a security audit.
What is not established
- The number of affected customers: not published beyond “a limited set”.
- The volume of files and which customers: undisclosed; one relayed account cites tens of thousands of files.
- Whether the stolen data has been used or sold: not stated.
- The identity of the criminal group: not named in the sources we read.
Summary table
| Point | What is stated as of November 7, 2026 | Source |
|---|---|---|
| First detection | October 6, 2026 | Cryptonews |
| Second intrusion | October 24 via a third-party cloud provider | GBHackers |
| Billing data | Addresses, invoices, payment history, balances, credit usage | Technology.org |
| Model data | Subset of incomplete weights and datasets, certain customers | Superpower Daily |
| Cards | Payment card numbers not stored or affected | The Neuron |
FAQ
Were complete model weights stolen?
Fireworks says a subset of incomplete model weights and datasets was accessed for certain customers; it has not reported that complete models were stolen.
Were payment card numbers exposed?
No. Fireworks says it does not store card numbers and they were not affected.
What is LLMjacking?
The theft and resale of access to AI models, typically by using stolen credentials or keys.
Method and limits
Sources checked on November 10, 2026: The Neuron, GBHackers, Technology.org, Superpower Daily and Cryptonews, read as search extracts. Fireworks’ own notice, Forbes and Bloomberg were not read directly. Not a security audit.
Sources
- The Neuron: Fireworks AI Was Hacked: How Criminals Profit From Stolen AI (verified 2026-11-08)
- GBHackers: Fireworks AI Hit by LLMJacking Cyberattack Exposing Customer Data and AI Model Weights (verified 2026-11-08)
- Technology.org: Fireworks AI Discloses Second Security Breach (verified 2026-11-08)
- Superpower Daily: Fireworks discloses wider breach impact, including customer datasets and unfinished models (verified 2026-11-08)
- Cryptonews: Fireworks AI says October security incident exposed customer data and model weights (verified 2026-11-08)
Spotted an error in this article? Report it · Our corrections policy