Zcash (ZEC) fell about 5.9% over 24 hours to roughly $1,190, moving lower alongside a broader crypto market decline that saw Bitcoin trade near $76,673 and total market capitalization drop about 1.8%.
Amid that decline, Helius founder Mert Mumtaz drew attention on X by linking Anthropic’s newly released FlashProve tool to Zcash’s pioneering privacy cryptography. Mumtaz said Anthropic and Claude are “now using the same cryptography that zcash was first in the world to pioneer,” adding “probably nothing.” The post gained more than 1,000 likes and over 170 replies at the time of reporting.
Mumtaz has previously argued that Zcash is undervalued by the market. In May, on the Bankless podcast, he said ZEC’s value should represent a share of Bitcoin’s total market value, with about 10% suggested as a possible comparison.
The Zcash network grew out of the 2014 paper Zerocash: Decentralized Payments from Bitcoin, which received a Test of Time Award in 2024 from the IEEE Symposium on Security and Privacy. The network launched in October 2016. Cryptographer Alessandro Chiesa, who co-wrote the Zerocash paper, is also linked to StarkWare and appears on the FlashProve contributor list alongside Anthropic’s Roy Rinberg and Alex Allain.
This is not the first Zcash-Anthropic overlap this year. Zcash founder Zooko Wilcox said in June that an audit by Anthropic’s Mythos AI, requested by Shielded Labs, found no serious new bugs in the protocol. Meanwhile, the Zcash community is debating how to handle its roughly $125 million grants treasury.
Anthropic released FlashProve on November 2 as an open-source research prototype. It uses a hash-based zkSNARK to prove that an answer came from a specific AI model without exposing the model’s weights. The tool is designed for mixture-of-experts models and works in two steps: first, the provider publishes a small commitment of about 6.5 MB, which takes about 39 minutes; second, for each prompt and reply, the system creates a short proof that the committed model produced those exact output tokens.
Anthropic tested FlashProve on Llama 4 Maverick, a Meta AI model with 400.7 billion parameters, using one NVIDIA H100 chip in compressed 4-bit form. The test handled about 110 tokens per second and processed 24 requests in roughly 223 seconds. Each proof was about 19 MB, and verification took only 17 to 21 seconds on a standard CPU. However, proof generation took about 72 to 94 times more GPU time compared with vLLM inference. Anthropic describes FlashProve as an intermediate research tool, not production software, and says it has not had an independent security audit.
