Ai model Anti-piracy mechanism
The method transforms neural networks into protected models using transformation and reverse functions, ensuring correct functionality only through authorized runtimes, effectively preventing piracy and unauthorized use.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- THALES DIS FRANCE SA
- Filing Date
- 2025-11-27
- Publication Date
- 2026-06-04
AI Technical Summary
Existing neural networks are vulnerable to piracy, where unauthorized parties create new models from licensed models, violating intellectual property rights, as legal remedies are insufficient to prevent such unauthorized use.
A method to transform unprotected neural networks into protected models using transformation functions, followed by a reverse function to ensure functionality, and employing a specialized runtime to access and interpret the protected models, incorporating techniques like altering descriptor tables, inserting incorrect values, and obfuscating operations.
Protects neural networks from unauthorized use by ensuring that unauthorized access results in incorrect outputs, thus preventing piracy while maintaining functionality for licensed users.
Smart Images
Figure EP2025084612_04062026_PF_FP_ABST