AI infrastructure provider Fireworks AI has disclosed two cybersecurity incidents in October 2026, highlighting growing threats to the artificial intelligence supply chain and the enterprises depending on third-party AI platforms.
Serving more than 10,000 customers, Fireworks provides infrastructure for deploying and running AI models. The incidents demonstrate how attackers increasingly target shared AI environments to access valuable corporate information and computing resources.
According to an investigation involving Palo Alto Networks' Unit 42, a financially motivated cybercriminal group compromised Fireworks systems, stealing account information, billing records, sensitive credentials, and AI-related datasets.
The exposure of secrets and authentication credentials creates additional risks because attackers could potentially exploit compromised access to penetrate customer environments, misuse APIs, or conduct further unauthorized activities.
A separate incident on October 23 reportedly involved a third-party cloud provider hosting Fireworks research and development machines. Attackers executed an unauthorized workload, resulting in further exposure of customer information and demonstrating vulnerabilities across interconnected infrastructure providers.
The consequences extended to downstream customers. Productivity platform Notion confirmed that some customers were affected, temporarily disabled its Insights feature, and discontinued serving open-weight AI models through Fireworks. The incidents also reflect concerns about LLMJacking, in which attackers hijack AI infrastructure or credentials to consume computing resources without authorization.
The broader warning is that AI supply-chain security must become a boardroom priority. Enterprises should strengthen API-key protection, credential rotation, workload isolation, cloud configuration monitoring, and third-party risk assessments. Continuous behavioral analytics, Zero Trust access controls, and real-time anomaly detection can help identify suspicious workloads before they escalate. As businesses increasingly depend on external AI infrastructure, protecting customer data, model assets, and computing resources requires security accountability across every technology provider in the ecosystem.















