FlashProve uses cryptographic proofs so users can confirm an output came from the model they think they are using.

Anthropic published an open source design for FlashProve, a verification system that produces a cryptographic proof tying a given output to a specific model. The proof acts as a mathematical guarantee, letting users confirm that the answer they received came from the model they believe they are using.

The system works like a cryptographic email signature, so recipients can confirm a message came from its claimed sender rather than an impostor. Anthropic researchers released a technical specification and working code under an open source license on GitHub, though they describe the work as intermediate progress rather than a finished research agenda.

Builders get a way to check that a vendor did not quietly swap in a cheaper model to cut costs or a stronger one to inflate benchmark results. Verification runs on a local device, and the fingerprints reveal no model weights, so commercial labs can publish them without exposing their most valuable asset.

Anthropic says it has no plans to deploy FlashProve at its own labs or anywhere else, and the team frames the release as a starting point. Watch whether outside researchers or model providers pick up the specification and build verification into their own release pipelines.

What matters

• Anthropic published FlashProve, an open source design for proving which model produced an output.
• Labs can publish fingerprints of model weights without exposing the private numbers behind commercial models.
• No lab plans to deploy FlashProve yet, and the researchers want others to take the code and improve it.

Why it matters

No lab plans to deploy FlashProve yet, and the researchers want others to take the code and improve it.

This GenAI News article was prepared in original wording using reporting and materials published by TechCrunch AI. Source reference: https://techcrunch.com/2026/11/03/anthropic-is-working-on-a-system-to-prove-ai-models-identity/.

Drafted by the GenAI News review pipeline.