News · SI
Fireworks AI says attackers reached some customer datasets and model weights
Inference provider Fireworks AI has expanded its October breach disclosure. Attackers reached billing and account data and, for a limited set of customers, secrets, datasets and incomplete model weights. Fireworks links the group to LLMjacking.
Fireworks AI, a popular inference and fine-tuning platform, has widened its disclosure of an October security incident. In a bulletin updated on 9 November, it says attackers reached account and billing information and, for a limited set of accounts, secrets, datasets and incomplete model weights[1].
01 / 03
Two routes in
Fireworks first reported unauthorised use of internal credentials on 9 October, when only stored secrets such as those used in evaluation jobs were known to have been accessed[3]. The new activity came later. A third-party cloud provider that hosted Fireworks R&D machines had its own incident on 23 October, and the next day the attacker used it to reach Fireworks development machines and run an unauthorised workload[1]. Fireworks says it detected that on the same day and hasn't seen further successful access to its customer platform since 25 October[1].
Bloomberg Law reports it is the company's second known hack in a matter of weeks, and that Fireworks is working with customers including Notion and Cursor on the incident[2].
02 / 03
What was and wasn't hit
Exposed data includes resource identifiers and descriptions, billing and shipping addresses, invoice and payment history, balances and credit usage. Fireworks says it doesn't store card numbers[1]. It has found no impact to Inference, Training SDK or Nexus, which are zero data retention by default, but it has paused self-serve dataset uploads and downloads for Batch Inference and dataset-based training while the investigation runs[1].
The company links the attack to a financially motivated group that sells stolen access to large language models through illicit routing gateways, known as LLMjacking, and is working with Palo Alto Networks Unit 42 on forensics[1].
03 / 03
Why it matters
Fine-tuned weights and training datasets are often a customer's most valuable AI asset, and they now sit with inference providers as well as in-house. Anyone who stored third-party keys with Fireworks should rotate them at the issuing provider, because deleting a stored secret doesn't revoke the underlying credential[3].
FAQ
Frequently asked questions.
01What data was exposed?
Resource metadata and billing information such as billing and shipping addresses, invoice and payment history, balances and credit usage. For a limited set of accounts, secrets, datasets and incomplete model weights were also accessed. Fireworks says it doesn't store payment card numbers.
02Were prompts or inference traffic affected?
Fireworks says it has not detected any impact to Inference, Training SDK or Nexus, which are zero data retention by default, though its investigation is ongoing.
03How did the attackers get in the second time?
A third-party cloud provider hosting Fireworks R&D machines had a security incident on 23 October. On 24 October the attacker used it to reach Fireworks development machines and run an unauthorised workload.
04Who is behind it?
Fireworks says the evidence links to a financially motivated cybercriminal group that sells stolen LLM access to illicit routing gateways, a practice known as LLMjacking.
05What should customers do?
Fireworks is contacting affected customers with account-specific steps. Its earlier guidance was to replace revoked API keys, rotate any third-party credentials stored with Fireworks at the issuing provider, and check for unrecognised keys, users, training jobs and datasets.
That’s the story
Building with what’s new?
Spectrum Web Co ships products, places talent, builds social narratives and builds custom business tools as monthly services.
Related stories