Anthropic Unveils FlashProve, an Open‑Source System to Cryptographically Verify AI Model Provenance
FlashProve: How the System Works and Why It Matters
Anthropic announced in a blog post published on Tuesday that it is developing FlashProve, an open‑source framework designed to deliver what researchers call “provable inference.” The system generates a cryptographic proof—a mathematical guarantee—that a particular output originated from a specific AI model, allowing users to confirm that the answer they received truly came from the model they intended to use.
The approach mirrors cryptographic email signatures, which let recipients verify that a message genuinely originates from its claimed sender. FlashProve is intended to be robust against impostors, even when the impersonator is the lab that built and operates the model. Researchers note that the technology could stop an AI provider from covertly substituting a cheaper model to reduce costs or from deploying a more expensive model to artificially boost performance metrics.
FlashProve could also address verification challenges for third‑party claims.
“A developer might claim that their software library was scanned for vulnerabilities by a frontier cybersecurity model, or a company might claim that its business plan was given the go‑ahead by an appropriate model,” the researchers write. “How can a user verify these claims?”
To function, the system requires AI laboratories to publish a “fingerprint” for each released model. These fingerprints can later be matched to specific outputs and verified cryptographically. Importantly, the fingerprints do not expose a model’s weights—the internal parameters that determine its behavior and that closed‑model creators such as Anthropic and OpenAI keep private. This design lets labs share fingerprints without surrendering their most valuable intellectual property, making the system compatible with commercial models as well. Additionally, the verification process is lightweight enough to run on a local device.
As part of the announcement, the research team released a detailed technical specification and functional code for various components of the system, all available under an open‑source license on GitHub. The work remains ongoing, and there are currently no plans to deploy FlashProve at Anthropic or any other laboratory.
“This research post is meant to share intermediate progress, and not a completed research agenda,” researchers explain in the announcement post. “We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it.”








