Anthropic Introduces FlashProve for Cryptographic AI Verification
Anthropic has shared an open-source system called FlashProve designed to cryptographically verify which specific AI model generated a given output.
Anthropic has released details regarding an open-source system named FlashProve, which aims to verify the origin of artificial intelligence outputs. According to a company blog post, the project provides what researchers term "provable inference." This concept relies on a cryptographic proof or mathematical guarantee to confirm that a response originated from a designated model rather than an imposter.
Functioning similarly to cryptographic email signatures, the proposed framework helps recipients validate the true sender of a message. The researchers explained that the system remains robust against attempts to disguise one model as another, even if the deception originates from the laboratory that built and operates the software. This transparency could prevent developers from swapping in cheaper alternatives to cut operating costs or utilizing higher-end systems to artificially inflate performance benchmarks.
Furthermore, the system addresses verification challenges raised by third-party claims. For instance, developers often assert that a software library underwent scanning by a frontier cybersecurity model, or businesses claim their plans were evaluated by advanced systems. FlashProve offers a standardized way to verify these assertions.
How the Technology Operates
The verification process requires AI developers to publish a unique "fingerprint" for every model release. These fingerprints can later be cross-referenced with specific outputs and authenticated cryptographically.
Crucially, these fingerprints do not expose the underlying model weights—the internal parameters dictating system behavior that commercial developers like Anthropic and OpenAI typically protect as proprietary assets. By keeping the weights private while sharing the fingerprint, the mechanism remains compatible with commercial models. Additionally, the architecture is lightweight enough to allow verification on local devices.
To encourage broader investigation, the team published a technical specification alongside working code on GitHub under an open-source license. Development is still ongoing, and the company clarified that there are currently no concrete plans to deploy the technology within Anthropic or other labs.
"This research post is meant to share intermediate progress, and not a completed research agenda," the researchers wrote. "We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it."
What it means for developers
As the ecosystem of available artificial intelligence options expands, validating backend infrastructure becomes increasingly important for software engineering teams. Developers can try top AI models cheaply through one API at https://apixoai.online. Systems like FlashProve highlight a growing industry focus on verification, integrity, and trust when integrating various machine learning models into third-party software libraries and enterprise applications.
Source: Anthropic is working on a system to prove AI models’ identity — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
