Fireworks AI Breach Widens to Expose Customer Billing Data and Incomplete Model Weights

On October 10, 2026, Fireworks AI announced that a security incident, initially reported last month, had exposed additional customer data, including billing information, secrets, datasets, and incomplete models. The company detected unauthorized use of internal Fireworks credentials on October 6, 2026, at 9 a.m. PDT, which allowed a third-party to access environment variables in limited customer accounts.

Initially, Fireworks AI believed only environment variables, such as API keys and tokens, were compromised, with no evidence of customer models, datasets, training data, inference traffic, or personal information being accessed. However, the company's latest bulletin revealed that the breach may have been more extensive than initially thought.

The attackers gained access to Fireworks' development environment via an unnamed third-party cloud provider on October 23, 2026, and then to Fireworks' development machines on October 24. On October 25, no further evidence of unauthorized access was detected on Fireworks' platform. Fireworks AI confirmed that payment card numbers, which are not stored by the company, were not impacted.

The company has implemented security hardening measures, including revoking and rotating compromised credentials, isolating affected systems, tightening access controls, and improving logging and monitoring systems. The incident is linked to a financially motivated cybercriminal threat actor group, which aims to sell tokens from unauthorized access to large language models (LLMs) to illicit routing gateways.

This activity is part of an emerging threat known as LLMjacking, where threat actors hijack a victim's LLM using stolen credentials to fuel criminal enterprises, leaving the legitimate owner to bear the costs.