Anthropic Launches Open-Source System FlashProve: Verifiable AI Model Identity via Cryptographic Fingerprints

On November 3, 2026, Anthropic's research team announced the launch of an open-source system called FlashProve, designed to provide cryptographic mathematical guarantees for the identity verification and output authenticity of artificial intelligence models.

The system aims to enable "verifiable inference," which can effectively prevent service providers from secretly replacing models in commercial operations (such as passing off a low-cost model as a high-performance model, or using a cheap model to exaggerate performance results). It can also be used to verify third-party claims about the execution of complex tasks.

Claude2, Anthropic, artificial intelligence, chatbot Claude

In terms of technical implementation, FlashProve requires AI labs to generate and publish specific "fingerprints" for the models they release, so that clients can perform cryptographic matching. The system's core advantage is that its fingerprinting mechanism does not reveal the model's internal weight data, balancing the confidentiality needs of commercial models with compatibility.

In addition, this lightweight system supports verification directly on local devices. Currently, Anthropic has open-sourced the project's detailed technical specifications and working code on GitHub. At this stage, the research is mainly intended for sharing milestone results and industry reference, with no specific commercial deployment plans for now.

This open-source advance provides an important technological evolution direction for addressing the trust and transparency pain points in large-model inference.

Putting Digital Fingerprints on AI Outputs: Anthropic Develops FlashProve to Leave Model Swapping Nowhere to Hide

Anthropic is developing an open-source verification system called FlashProve, aimed at addressing whether the AI answers users receive are truly generated by the model they claim to be. In a Tuesday blog post, the company said its 'provable inference' will use cryptography to provide a mathematical guarantee for every output, confirming that it came from a specified model, similar to how cryptographic email signatures verify the sender. The system can prevent model A from masquerading as model B, close a hidden backdoor for cost cheating, and expose inflated benchmark results, striking directly at a long-standing unspoken practice in the industry.

Anthropic's Killer Verification System FlashProve Officially Open-Sourced: Precisely Exposes Rebranded Models in 17 Seconds, Verification Speed Soars 5,000x

The prosperity of the open-source AI ecosystem has given rise to the chaos of 'rebranded models': some developers slightly repackage open-source models, swap weights, or tamper with outputs, then claim them as self-developed new models. To tackle this chaos at its root, leading AI company Anthropic has launched and open-sourced FlashProve, a provable inference system for large models. Relying on underlying cryptographic technology, the system can accurately verify a model's true identity and curb rebranding fraud.

Claude Issues Itself an ID Card: Anthropic Uses Zero-Knowledge Proofs to Cut Off the Gray Market of Model Wrapping

In the large-model supply chain, there is a gray area where users pay for top-tier configurations but run downgraded models. Anthropic, together with zero-knowledge proof researchers, has open-sourced FlashProve: before launch, a single H100 generates and publishes a "weight fingerprint" ID in about 39 minutes; after each conversation, the GPU generates a proof of about 19 MB; users can run cp_verify on an ordinary CPU to verify authenticity in 17 seconds, without needing weights or a GPU. The underlying zkSNARK uses about 2,000 sumchecks per request, with security relying on BLAKE2. AI writes code, Lea

Anthropic Launches Open-Source System FlashProve: Verifiable AI Model Identity via Cryptographic Fingerprints

Related Recommendations

Using Its Own Large Model to Build Its Own Chips: Anthropic Forms 15-Person Chip Design Team, Claude Helps Accelerate R&D

Putting Digital Fingerprints on AI Outputs: Anthropic Develops FlashProve to Leave Model Swapping Nowhere to Hide

Anthropic's Killer Verification System FlashProve Officially Open-Sourced: Precisely Exposes Rebranded Models in 17 Seconds, Verification Speed Soars 5,000x

Claude Issues Itself an ID Card: Anthropic Uses Zero-Knowledge Proofs to Cut Off the Gray Market of Model Wrapping

Anthropic product team sees personnel change, Claude Code lead departs