Anthropic is developing an open-source system to verify which AI model generated a specific output, as detailed in a blog post released on Tuesday. Named FlashProve, this system aims to deliver "provable inference," which establishes a cryptographic proof—a mathematical guarantee—that a specific result originated from a designated model. This functionality allows users to ensure the answer they are receiving genuinely comes from the model they believe they are using.
Resembling cryptographic email signatures that authenticate a sender, FlashProve is designed to withstand spoofing attempts, even from the laboratory that constructed the model. Specifically, the researchers emphasize that this tool could prevent an AI company from covertly swapping out a cheaper model to reduce costs or a more advanced model to exaggerate performance metrics.
Beyond internal verification, FlashProve could validate third-party assertions regarding specific tasks. "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 noted. "How can a user verify these claims?"
To function, the system requires AI laboratories to publish a "fingerprint" for each released model, which can subsequently be matched to outputs for cryptographic verification. A critical advantage is that these fingerprints do not expose the model's weights—the internal parameters defining behavior—which are typically guarded by closed-source companies like Anthropic and OpenAI. Consequently, labs can publish these identifiers without compromising their most valuable IP, making the system compatible with commercial models. Additionally, the architecture is lightweight, allowing verification to occur locally on a device.
As part of the announcement, the researchers released a comprehensive technical specification and functional code for various system components. This code is publicly available under an open-source license on GitHub. The research is currently ongoing, with no immediate plans for deployment at Anthropic or other labs.
This research post is meant to share intermediate progress, and not a completed research agenda," researchers explained in the announcement. "We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it."
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.