Wire flash
Anthropic open-sources FlashProve to mathematically prove which AI model generated a reply
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
Anthropic has open-sourced FlashProve, a system that lets a server mathematically prove which AI model generated a given reply, with verification checkable on an ordinary CPU. The tool addresses cases such as resellers quietly serving a cheaper model than the one a customer paid for. Anthropic tested FlashProve on Meta's Llama 4 Maverick, a 400B-parameter open model, where a single Nvidia H100 GPU produced proofs at 110 tokens per second, each about 19MB. A regular CPU checks each proof in about 17 seconds, compared with VerInf, the only earlier system known to prove this model, which produced 93.6GB proofs that took almost a day to check. Claude built the system, and because its code is hard to audit, the security guarantees are mathematically proved in Lean 4, a proof-checking language.
Source report
Anthropic has just open-sourced FlashProve, a system that lets you check which AI model actually wrote a reply — and it's checkable on an ordinary CPU.
Why It Matters
For example, if you pay for a top AI model through a reseller, you currently can't check whether it quietly served a cheaper one instead. Anthropic's FlashProve lets the server prove mathematically which model answered.
Performance and Testing
Anthropic tested FlashProve on Meta's Llama 4 Maverick, a large 400B-parameter open model:
- A single Nvidia H100 GPU produced proofs at 110 tokens/second, each about 19MB.
- A regular CPU checks each proof in about 17 seconds.
- By comparison, VerInf — the only earlier system known to prove this model — produced 93.6GB proofs that took almost a day to check.
Security Guarantees
Claude built the system, and because its code is hard to audit, the security guarantees are mathematically proved in Lean 4, a proof-checking language.
Source
rohanpaul_aiNeutral / independent
Part of this Story
Anthropic Open-Sources FlashProve to Cryptographically Verify Which AI Model Produced a Response