Anthropic unveils FlashProve, a cryptographic way to prove which model answered

Anthropic published a research blog post on Tuesday describing FlashProve, an open-source system for verifying which AI model produced a given output. According to TechCrunch, the system is built around what the researchers call "provable inference" — a cryptographic proof, essentially a mathematical guarantee, that a response came from a specific model.

The problem it targets is trust in the model behind an API call. Today a user has little choice but to take a provider's word for which model answered; FlashProve is meant to turn that claim into evidence the user can check, much as cryptographic email signatures let a recipient confirm who sent a message.

The design is meant to hold up against impostors — including the lab that built and operates the model. The researchers point to two concrete failure modes: an AI company quietly swapping in a cheaper model to cut costs, or a more expensive one to inflate performance results. FlashProve is intended to leave a detectable trace when that happens.

It could also be used to check third-party claims about which model did a piece of work. "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?"

Making it work requires labs to publish a "fingerprint" for each model they release, which can later be matched against specific outputs and verified cryptographically. Crucially, the fingerprints do not reveal a model's weights — the internal numbers that define its behavior and that closed-model makers keep private — so labs could publish them without giving up their most valuable asset, which opens the scheme to commercial models.

The system is also designed to be lightweight, so verification can run on a local device. Alongside the announcement, the researchers released a detailed technical specification and working code for various components under an open-source license on GitHub. Research is ongoing, and there are no plans to deploy FlashProve at Anthropic or any other lab.

The researchers describe the post as intermediate progress rather than a finished research agenda. "We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it," they write. The broader signal is that model identity and output provenance are becoming infrastructure concerns: once models are embedded in code audits, compliance checks and business decisions, which model answered stops being a technical footnote and becomes a question of accountability. What to watch next is whether any lab actually publishes fingerprints, and whether lightweight verification holds up in real products.