Fireworks AI, which runs a platform for running and training AI models and is backed among others by Nvidia, expanded its disclosure of an October security incident on November 9. According to the company, attackers reached its development machines through a breached cloud provider and read customers' account and billing details. For a limited group of customers, they also reached secrets (such as access keys), datasets and partial model weights.
The company stresses no impact was detected on its inference services, training SDK or Nexus product, which by default operate without retaining data. It also does not store card numbers. In response, Fireworks replaced keys and service accounts, removed unnecessary elevated privileges, isolated systems and tightened access to data stores. The investigation is being run by, among others, Palo Alto's Unit 42, and law enforcement has been notified. According to Bloomberg, it is the company's second breach in weeks.
Fireworks is not alone. In a report published by cyber company Wiz at the end of October, titled 'Black October,' at least nine AI development platforms were identified that reported security incidents that month, including Lightning AI, Railway, Warp and Northflank. According to Wiz, no single vulnerability links the cases, and it has not been shown that the same attacker is behind them. But some attackers were after one thing: environment variables containing access keys for AI services.
The reason is simple: an AI service API key is money. Whoever steals it can run expensive models on the victim's account, sell the access to others or use it for purposes the provider would block. Northflank confirmed customers reported unauthorized use of AI keys stored on the platform. According to Wiz, in one case examined, stolen keys for Amazon's Bedrock service were used for hundreds of thousands of calls.
The lesson for companies, including many Israeli startups building on these platforms: the new AI platforms grew fast, much faster than their security infrastructure. They hold customers' most sensitive assets, data, models and keys, and so have become a preferred target. Anyone using them should assume they will be breached one day, and plan accordingly: limit every key, monitor its use and rotate it regularly.
