Anthropic Introduces Cryptographic Verification Method for AI Answers
Anthropic, the developer of the Claude AI language model, has released a study on FlashProve. The work utilized zkSNARK technology for the cryptographic verification of artificial intelligence responses.
The research was conducted jointly with Alessandro Chiesa, an expert at StarkWare. The technology allows confirming the authenticity of a response without revealing the internal parameters of the model.
Key results of the test include the following metrics:
- Acceleration of cryptographic proof generation up to 110 tokens per second
- Application to an AI model with 400 billion parameters
- Verification of the answer source without leaking data about the model
zkSNARK technology is also used in scaling solutions for Starknet and Mantle. However, Anthropic has not announced any direct collaboration with Starknet or the use of the STRK token.









