Model Weight Exposure

Model weight exposure refers to the unauthorized copying, leaking, or theft of the trained numerical parameters that define an AI model, which is the core asset a developer spends heavily to create. For enterprises, a leak can hand a competitor or malicious actor a working copy of proprietary capabilities, and it can expose training data or sensitive behavior embedded in the model. Governance teams need to track how weights are stored, who can access them, and what contractual and regulatory obligations apply if they escape.

1 item

Corporate PolicyUS2026-11-10

Fireworks AI's Second Breach Exposes Model Weights, Billing Data via Cloud Provider

Fireworks AI disclosed a second cyberattack in October 2026, tracing to a compromised third-party cloud provider. The breach exposed customer billing records, partial model weights, and datasets for certain customers. Customers including Notion, Cursor, and Harvey were affected.

Fireworks AI LLMJacking third-party risk AI security breach model weight exposure data breach supply chain risk