Zcash (ZEC), a leading privacy-focused cryptocurrency, dropped approximately 5.9% within 24 hours, reaching around $1,190. The decline came as the broader crypto market experienced losses, with Bitcoin (BTC) trading near $76,673 and the total market value down by 1.8%.
Anthropic’s FlashProve draws attention to Zcash’s cryptography
Mert Mumtaz, founder of Helius, highlighted that Anthropic’s new FlashProve prototype uses the same category of privacy cryptography first introduced by Zcash. Anthropic, an artificial intelligence company, unveiled FlashProve as an open-source research tool that leverages zero-knowledge proofs to confirm outputs from AI models without revealing their internal data.
Mumtaz posted on X that “anthropic/claude are now using the same cryptography that zcash was first in the world to pioneer,” and later remarked “probably nothing.” The discussion quickly drew significant engagement from the crypto community, with over 1,000 likes and more than 170 replies.
Anthropic’s FlashProve employs privacy math pioneered by Zcash, demonstrating the influence of cryptographic research from the crypto sector on advancements in artificial intelligence verification.
Earlier this year, Zcash founder Zooko Wilcox reported that an audit of the protocol by Anthropic’s Mythos AI, commissioned by Shielded Labs, did not uncover any serious new bugs. This marks another instance of overlap between Anthropic and Zcash in 2024.
Mini dictionary: Anthropic is a US-based artificial intelligence company known for developing advanced AI models and safety-focused research projects. FlashProve is their open-source prototype designed to provide cryptographic proof that specific AI model outputs originated from the claimed source model, without revealing sensitive internal data.
Legacy of Zcash’s cryptographic research
Zcash is built on research from the 2014 Zerocash paper, which introduced practical privacy to decentralized payments through zk-SNARKs (zero-knowledge succinct non-interactive arguments of knowledge). The Zcash network was officially launched in October 2016, and in 2024, the protocol received the IEEE Symposium on Security and Privacy’s Test of Time Award for its groundbreaking impact.
Alessandro Chiesa, a co-author of the original Zerocash paper and currently a professor at EPFL, is connected to both Zcash and FlashProve. Chiesa is also recognized for his work with StarkWare and appears alongside Anthropic’s Roy Rinberg and Alex Allain in FlashProve’s contributors list.
Technical performance and limitations of FlashProve
Anthropic’s FlashProve relies on hash-based zero-knowledge proofs to ensure response authenticity in AI systems. The tool operates in two steps: the model provider first publishes a commitment locking in one set of weights (about 6.5 MB) in roughly 39 minutes, then generates short cryptographic proofs for each prompt and output token pair.
In benchmarking, Anthropic tested FlashProve on Llama 4 Maverick, a large language model by Meta, comprising 400.7 billion parameters, compressed to a 4-bit version. Using Nvidia’s H100 GPU, the prototype processed about 110 tokens per second. The system completed 24 requests, each with 1,024 tokens, in about 223 seconds, producing proofs around 19 MB apiece. Verification of each proof required only 17–21 seconds on a standard CPU and did not expose the underlying weights of the AI model.
However, FlashProve’s proof generation currently requires around 72 to 94 times longer per GPU compared to vLLM inference, depending on load. Anthropic described FlashProve as an intermediate research project, making it clear that the tool is not production-ready and has not undergone an independent security audit.
| Feature | FlashProve | vLLM Inference |
|---|---|---|
| Proof generation time (per GPU) | 72–94x slower | 1x (baseline) |
| Verification time (CPU) | 17–21 sec per proof | Not applicable |
| Security audit | None completed | N/A |
Zcash community debates governance and funding
The Zcash community is currently engaged in discussions regarding management of its grants treasury, which amounts to approximately $125 million. This debate reflects broader questions about the future direction and governance of privacy-centric blockchain projects.
