Anthropic is developing FlashProve — a system for cryptographically verifying which AI model produced an answer

What Anthropic presented

The company Anthropic is working on an open system that makes it possible to verify which AI model produced a given output. The system is called FlashProve and is described as “provable inference” — a cryptographic proof, essentially a mathematical guarantee that a specific answer came from a specific model. This was reported by TechCrunch citing a blog post published on Tuesday, November 3, 2026.

Why it matters

The principle is comparable to cryptographic email signatures, by which a recipient confirms that a message really came from the claimed sender: FlashProve is meant to be resistant to attempts to pass off one model as another — even if the substitution is carried out by the very lab that created and maintains the model. The researchers note that the system could prevent an AI company from secretly swapping in a cheaper model to save money or, conversely, a more expensive one to inflate quality metrics. Another scenario is verifying third-party claims: as the authors write, a developer might claim that its library was audited for vulnerabilities by a leading cybersecurity model, or a company might claim that its business plan was approved by a suitable model; the question is how a user could verify this.

How it works

For the system to work, AI labs would have to publish a “fingerprint” of each released model, which is then matched against specific outputs and verified cryptographically. The key point: fingerprints do not reveal the model’s weights — the internal numbers that determine its behavior, which the creators of closed models such as Anthropic and OpenAI keep secret. This allows labs to publish fingerprints without giving away their main asset, meaning the system is compatible with commercial models as well. By design, verification is intended to be light enough to run on a local device.

Project status

Along with the announcement, the researchers published a detailed technical specification and working code for individual components of the system, available under an open license on GitHub. That said, this is strictly a research project: there are no plans yet to deploy FlashProve at Anthropic or any other lab. “This research post is intended to share intermediate progress, not a completed research program,” the authors explain. “We are sharing this with the explicit goal that others look at what we have done, take it, and improve on it.” Thus, at the current stage this is about intermediate results and an open starting point for third-party developers, not a finished product or a decision to adopt it.