Fireworks AI Discloses Follow-Up Breach After Third-Party Cloud Provider Hack Exposes Customer Metadata and Partial Model Weights

Fireworks AI has reported a second security incident in October 2026 after a third-party cloud provider was breached, allowing attackers to reach the company's development machines. The threat actors accessed resource metadata, billing information, and incomplete model weights and datasets belonging to some customers. No customer payment card numbers were exposed because Fireworks does not store them. The company linked the activity to a financially motivated cybercriminal group focused on LLMjacking, the unauthorized use of stolen API keys to run large language models through illicit gateways. Fireworks responded by revoking credentials, isolating systems, and engaging Palo Alto Networks Unit 42 for further investigation. Earlier in the month the same group had obtained environment variables containing API keys and tokens from a limited number of accounts.

BleepingComputerTranslated from English

Read at source

Fireworks AI has disclosed a new security breach in which attackers accessed its development machines after compromising a third-party cloud provider.

Fireworks provides a high-performance inference platform that lets developers deploy and run open-source artificial intelligence models such as Mistral, Meta's Llama, and Google's Gemma without managing hardware or complex infrastructure.

The company had already notified law enforcement and customers of an earlier incident in early October after attackers obtained environment variables that contained API keys and tokens stored as secrets for a limited number of accounts. Evidence at the time pointed to a financially motivated cybercriminal group engaged in LLMJacking, the sale of stolen tokens for unauthorized access to large language models through illicit routing gateways.

After revoking the compromised credentials, Fireworks discovered a second incident on October 23. In this breach the threat actors accessed resource metadata including identifiers, display names, descriptions and creation times, as well as billing information such as addresses, invoice history, account balances and credit usage. They also obtained incomplete model weights and datasets for some customers. The company confirmed that attackers could not reach customer payment card numbers because Fireworks does not store them.

The root cause was a security incident at a third-party cloud provider that hosted Fireworks R&D machines on October 23. The unauthorized party used that foothold the following day to reach internal development systems and run an unauthorized workload that led to additional customer data exposure.

Fireworks detected the activity the same day and immediately applied remediation steps. As of October 25 the company had not observed further successful unauthorized access on the customer-facing platform. Remediation actions included revoking and replacing credentials and service accounts, restricting unauthorized network pathways, isolating affected systems, and tightening access controls on data storage and analytics systems.

The company is now working with Palo Alto Networks Unit 42 to strengthen defenses and complete a full investigation of the incident's scope and impact.

Meridian digestThe week's most consequential security stories, distilled into one briefing.See the digest

More in Breaches & Leaks

SEGA Newsletter Data Breach Exposes 2.2 Million Email Addresses via Third-Party Provider ARARA
York University Breach Exposes Personal Information Submitted Since 2007
Binance Denies Alleged Data Breach Claims Involving 335 Million User Records by ShinyHunters Impersonator