Fireworks AI breach exposes customer data

The Fireworks AI breach exposed customer information after attackers compromised a third-party cloud provider and reached the company’s development machines. In a November 9 advisory, the AI infrastructure startup said the incident affected resource metadata, billing information and incomplete model weights and datasets belonging to some customers.

The disclosure follows a separate security incident announced in early October involving API keys and tokens. However, the latest advisory describes additional customer data exposure through the cloud provider that hosted Fireworks research and development machines.

Attackers reached development machines through provider

According to Fireworks, the third-party cloud provider experienced a security incident on October 23. The following day, attackers used that access to reach Fireworks development machines and run an unauthorized workload.

Fireworks said it detected the activity on October 24 and immediately began remediation. Its response included closing known critical vulnerabilities and taking steps to contain the intrusion.

The company reported no further evidence of successful unauthorized access to its customer-facing platform from October 25 onward. Nevertheless, it continues to investigate the full scope and impact of the incident.

Fireworks operates an inference platform that helps developers deploy and run open-source AI models. Its services support models such as Mistral, Meta’s Llama and Google’s Gemma while handling the underlying infrastructure.

Fireworks AI breach affected billing and model data

The attackers accessed resource metadata, including identifiers, display names, descriptions and creation times. They also reached billing information such as billing and shipping addresses, invoice and payment histories, account balances and credit usage.

Additionally, the exposed material included incomplete model weights and datasets for some customers. The disclosure does not establish that attackers obtained complete copies of those customers’ models or datasets.

Payment card numbers remained outside the breach’s scope because Fireworks does not store them. However, the company has not publicly provided a total number of affected customers in the supplied advisory details.

Company replaces credentials and restricts access

Following the attack, Fireworks revoked and replaced credentials and service accounts. It also isolated affected systems, restricted unauthorized network pathways and tightened access to data storage and analytics systems.

Meanwhile, the company is working with external incident-response specialists to investigate the intrusion and strengthen its defenses. That investigation aims to establish the full extent of the customer data exposure.

The earlier October incident involved environment variables containing API keys and tokens stored as secrets for a limited number of accounts. In response, Fireworks revoked compromised internal credentials, affected customer API keys and keys that the attackers had created.

Fireworks linked that earlier activity to financially motivated criminals seeking to resell unauthorized access to large language models. This practice, known as LLMjacking, uses stolen credentials to generate AI workloads at the victim’s expense.

The latest Fireworks AI breach disclosure adds billing records and customer model material to the information affected across the two incidents. However, the company’s assessment of the attackers’ financial motives concerned the earlier credential-related intrusion.