Fireworks, an AI infrastructure provider founded in 2022 with over 10,000 customers, revealed that attackers stole billing information and AI-specific data during two breaches last month. The company, which supplies AI systems and tools to enterprises, published an updated disclosure on Monday after engaging Palo Alto Networks Unit 42 to investigate the initial breach discovered on October 6.
Investigators determined that a financially motivated cybercriminal group carried out the attack, obtaining account credentials, billing records, and secrets—confidential digital credentials that enable access for users or applications. Fireworks stated the breach affected a "limited number of customer accounts," with all impacted parties notified.
A second incident occurred on October 23 when attackers compromised a third-party cloud provider hosting Fireworks' research and development infrastructure. "On October 24, the unauthorized third party used [the provider] to reach Fireworks development machines and then run an unauthorized workload that led to additional customer data impact," the company stated. The breach exposed internal data, billing records, invoices, payment history, datasets and other AI-related information. Fireworks is collaborating with Palo Alto Networks to strengthen security defenses and revoke compromised credentials.
Notion, a digital workspace platform, disclosed that some of its customers were affected by the Fireworks incidents. Notified on November 2, Notion disconnected from Fireworks and purged all data stored in the system. "On November 6, Fireworks notified us that malicious actors may have accessed potential customer data. Our team immediately began another internal investigation," the company said. Notion had relied on Fireworks for classification tasks supporting its Insights feature, which analyzes customer AI usage, and as a provider for open-weight models. "The Insights feature has been disabled as we investigate this incident, and we are no longer serving open-weight models through Fireworks."
LLMJacking
Fireworks' breaches fit within a broader pattern of attacks that Google's threat researchers recently highlighted. The Google Threat Intelligence Group (GTIG) and Mandiant published analysis last week on a technique called "LLMJacking," in which attackers seize an organization's credentials for large language model access.
Mandiant has responded to multiple LLMJacking incidents throughout 2026 affecting various cloud and LLM providers across organizations in North America, South America, and Asia Pacific in the financial services, retail, technology, and transportation sectors. "Unlike the gradual resource consumption associated with cryptojacking, LLMJacking can generate large spikes in fraudulent cloud charges in a matter of hours," the researchers noted. "Across individual incidents, we observed fraudulent charges associated with resource consumption exceeding $100,000 in a single day, with cumulative charges at some organizations reaching $1 million before access was revoked."
GTIG has identified a "diverse set of financially motivated threat clusters operating out of multiple regions globally" orchestrating these campaigns. The threat actors are becoming increasingly sophisticated, with some groups specializing in initial access before transferring compromised credentials to other groups that monetize and resell unauthorized LLM access.
Investigations revealed that threat groups primarily target tokens and credentials that provide entry points into enterprise infrastructure. "As enterprises integrate higher-tier models and agentic workflows into production environments, cybercriminals will likely refine automated reconnaissance to target exposed API tokens and exploit vulnerabilities in model routing middleware," GTIG stated. "This operational specialization will sustain a growing illicit resale marketplace, allowing threat actors to rapidly siphon enterprise quotas while passing the billing costs to victims before compromises are detected."



