Anthropic found a way to prove which AI model answered you — but there’s a costly catch
FlashProve uses zkSNARKs technology to cryptographically prove that a response came from a specific set of model weights while keeping those weights hidden. The system addresses a real developer concern: AI API providers could theoretically route requests to cheaper models or attackers could compromise responses without detection. Although Anthropic demonstrated FlashProof on Meta's Llama model and released the code on GitHub, it remains a research prototype with impractical proof-generation costs for routine API usage. The company has not yet integrated it into the Claude API.
Key points
- FlashProve uses zkSNARKs to verify model identity without revealing weights
- Addresses developer concerns about undetected model switching or compromised inference
- Open-source but remains experimental with prohibitive costs for practical API use
- Not yet integrated into commercial Claude API offerings
Summary and key points written by AI from the headline and a short excerpt, so they can be incomplete or wrong. The New Stack did the reporting; read the original for the full story.
Read the full story at The New Stack
