Anthropic Releases FlashProve: Cryptographic Proofs That a Reply Came From Committed Weights
MarkTechPosten
Anthropic released FlashProve, an Apache 2.0 zkSNARK prototype that proves a reply came from a committed model without revealing weights. It proves Llama 4 Maverick at 110 tokens/second on 1 H100, with 19 MB proofs. The post Anthropic Releases FlashProve: Cryptographic Proofs That a Reply Came From Committed Weights appeared first on MarkTechPost .
This is a short summary published by AI Global Wire. The full article is owned and hosted by MarkTechPost — open it there to read it in full.
Read the full story at MarkTechPost- Anthropic
- Meta
- Verktyg
Related AI news
- OpenAI taps former Uber executive Jill Hazelbaker to lead corporate affairsDIGITIMES · November 4, 2026
- Meta launches $15b bond sale to fund AI infrastructureTech in Asia · November 4, 2026
- AMD beats expectations as agentic AI sparks its data center revenue growth enginesSiliconANGLE · November 4, 2026
- Tech giants commit over US$3 trillion as AI infrastructure faces funding, power testDIGITIMES · November 4, 2026
- Google offers weak response to accusations it would hold Search hostage if schools restrict AI toolsSiliconANGLE · November 4, 2026
- Google offers tepid response to accusations it would hold Search hostage if schools restrict AI toolsSiliconANGLE · November 4, 2026