ðŸ‡ðŸ‡· Croatian, 5 November 2026, AI verification
Anthropic Develops FlashProve System for Verifying AI-Generated Content
The system provides mathematical guarantees that a specific output comes from a known AI model.
Anthropic researchers are developing an open-source system called FlashProve to verify whether a specific output was generated by a known AI model. The system is designed to provide a cryptographic proof, or mathematical guarantee, that the output comes from the intended model. This can help prevent companies from secretly replacing high-end AI models with cheaper alternatives or manipulating performance results.
The system works by allowing labs to publish public fingerprints of each model they deploy, which can then be compared to the output and verified using cryptography. This approach protects business secrets and makes it easy to verify outputs directly on a user’s local device.
The project is still in its early stages, and researchers have published a detailed technical specification and functional code on GitHub under an open-source license. However, there are currently no plans for its implementation within Anthropic or other AI labs.