Fireworks, a startup that runs and trains open-source AI models for developers, said a recent hack exposed customers’ billing information, model weights and datasets, and linked the attack to a criminal group that sells stolen access to AI tokens. The disclosure updates an October incident in which an intruder used internal credentials to reach model evaluation data, when Fireworks said it saw no sign that customer models or data were touched. Fireworks has told large customers including Harvey, which said none of its customer data was affected, and some that spend millions a year are weighing other providers, according to The Information. Fireworks had been set to begin raising money this week at a valuation of $30B or more, up from $17.5B in July.
