The Decoder· Maximilian Schreiner·· 11 hours agoSelectedAI score65
Anthropic open-sources FlashProve, an inference verification prototype that verifies model output provenance without exposing weights
Anthropic's new prototype proves which model wrote an AI response without exposing its weights
AI guide
Anthropic has released and open-sourced FlashProve, a research prototype that uses zk-SNARKs to verify whether AI-generated content genuinely originates from a claimed model without exposing its weights.
Why it matters
The approach demonstrates a technical pathway to verify LLM inference provenance without relying on trusted hardware, providing a practical, purely cryptographic reference for model auditing, regulatory oversight, and tamper resistance.
Source:The Decoder · the-decoder.com