Fireworks AI breaches expose billing data and LLM tokens to financially motivated LLMJacking groups

Fireworks, founded in 2022 and serving over 10,000 customers, initially detected compromise on October 6. Palo Alto Networks Unit 42 traced activity to a known financially motivated cluster that exfiltrated account details, payment history, and secrets used for model inference. A second breach on October 23 originated from a compromised cloud provider hosting development workloads, allowing unauthorized inference jobs and further data access.

Mandiant and Google Threat Intelligence documented parallel LLMJacking campaigns in 2024 where groups target enterprise LLM credentials to generate fraudulent charges exceeding $100,000 daily and cumulative losses near $1 million. Technical indicators show credential theft as the primary objective, with some clusters selling access rather than consuming resources directly. Fireworks matches this pattern exactly.

Notion disconnected its systems on November 2 after notification and disabled its Insights feature that routed classification jobs through Fireworks. The incidents highlight how AI inference platforms concentrate high-value tokens that enable rapid monetization without the slow burn of cryptojacking.

Fireworks is revoking credentials and hardening posture with Unit 42. Expect similar disclosures from other inference providers as billing anomalies surface in procurement and incident logs.