Progress on provable inference for large language models
What happened
Anthropic has shared progress on a system called FlashProve that uses zkSNARKs to cryptographically prove that AI responses come from a specific model. For a large 400B-parameter model, the system generates proofs at around 110 tokens per second on a single H100 GPU, which is significantly faster than prior open-source systems.
Why it matters
This technology allows users to verify that responses are generated by the claimed model rather than a substituted or tampered version, potentially enabling secure applications and helping verify compliance with AI pacing requirements.
Is this a big deal?
Get stories like this every morning
The day's AI news in plain English, in your inbox at 8:30 IST. Free.
More from Anthropic
Anthropic Upgrades Claude Code Projects to Run Parallel AI Workers Hands-FreeAnthropic · 14 Nov 2026Claude designed FPGA scaler board for 6th-gen consolesAnthropic · 13 Nov 2026I asked Claude Fable 5.2 to turn my site into a 90s Silicon Graphics workstationAnthropic · 13 Nov 2026Anthropic is endorsing SB 53Anthropic · 13 Nov 2026Understanding and addressing AI harmsAnthropic · 13 Nov 2026This page was collected from anthropic.com and summarised automatically by AIPulsee's pipeline. See how it runs.