By Hamza Faisal

Fireworks AI has sharply widened its account of an October security incident, saying hackers reached customer billing records and, for a limited set of accounts, datasets and incomplete model weights. The disclosure, published on November 9, describes a second intrusion that arrived weeks after the first, through a cloud provider hosting the company's research machines.

The AI inference platform, which serves more than 10,000 enterprise customers including Uber, Shopify and Cursor, first detected unauthorised use of internal credentials on October 6 and disclosed the incident on October 9. At that stage it said it had found no unauthorised access to customer models, datasets, training data, personal information or inference traffic. The new update says that picture was incomplete.

A second way in through a cloud provider

According to the updated security bulletin, a third-party cloud provider hosting Fireworks AI's research and development machines suffered its own security incident on October 23. The next day, the attacker used that access to reach Fireworks' development machines and run an unauthorised workload, which led to further customer data being exposed. Fireworks said it detected the activity the same day and took remediation steps, including closing known critical vulnerabilities.

The company says it has detected no further successful unauthorised access on its customer platform since October 25. It also says the core services were untouched: it found no detected impact on Inference, its Training SDK or Nexus, which it describes as zero data retention by default. As a precaution, it has disabled self-serve dataset uploads and downloads used by Batch Inference and dataset-based training while the investigation continues.

What was exposed, and what was not

The exposed information includes account and billing records — billing and shipping addresses, invoice and payment history, account balances and credit usage — plus resource metadata such as identifiers, display names and creation times. For a limited group of customers, the attackers also reached secrets, datasets and a subset of incomplete model weights, the numerical values that make up a trained model.

Payment card numbers were not affected, the company said, because it does not store them. Affected customers are being notified with account-specific details and recommended response steps. Fireworks said it has notified law enforcement and brought in Palo Alto Networks' Unit 42 for forensic analysis and security hardening.

Customers are already walking away

Fireworks attributes the attacks to a financially motivated cybercriminal group and calls the motive "LLMjacking" — stealing access to large language models so it can be resold through illicit routing gateways. Google's Threat Intelligence Group and Mandiant published research on the rise of LLMjacking on November 2, days before Fireworks' expanded disclosure.

The commercial fallout is already visible. Two customers told Forbes they had substantially moved their business off Fireworks' systems, and Bloomberg described the November 9 notification as the company's second hack disclosure in a matter of weeks. The incident also lands in a wider pattern: security firm Wiz reported on October 30 that at least nine AI developer platforms disclosed security incidents during October, most involving attackers hunting for stored credentials and environment variables that grant access to AI models.

By Hamza Faisal