AEGLIX News
Fireworks AI revealed a fresh security compromise on October 23, explaining that intruders reached its development machines by first breaching a third‑party cloud provider. The new incident comes after an earlier intrusion in early October when threat actors accessed environment variables that held API keys and tokens stored as secrets for a limited set of accounts. Fireworks said it uncovered the latest breach after it revoked compromised internal credentials, affected API keys, and keys the attackers themselves had created.
According to Fireworks, the newly disclosed breach gave threat actors access to resource metadata, billing details, and incomplete model weights and datasets belonging to some customers. The company stressed that payment card numbers were not exposed because it does not store them. As of October 25, Fireworks had not found additional evidence of successful unauthorized access by the threat actor on the customer‑facing platform.
The episode underscores the rising danger of supply‑chain attacks aimed at AI infrastructure. Google's Threat Intelligence Group and Mandiant recently reported that LLMjacking campaigns are pushing costs to record levels for compromised organizations, with cloud AI charges climbing as high as $1 million from fraudulent calls made with stolen API keys. Fireworks' disclosure shows the attackers moved from the breached cloud provider into the company's development environment, enabling access to sensitive customer artifacts such as model weights and datasets.
Following the attack, Fireworks revoked and replaced credentials and service accounts, blocked unauthorized network paths, isolated affected systems, and tightened controls on data storage and analytics systems. The firm is collaborating with Palo Alto Networks' Unit 42 incident response division to secure its environment and investigate the full scope and impact of the incident.
Report prepared 2026-11-10 · Based on reporting from BleepingComputer. Source publication date above; subsequent developments may not be reflected.
Sources
Read original source · BleepingComputer (opens in a new tab)