- Published on
Anthropic is working on a system to prove AI models’ identity
- Authors
- Name
- DeepHorus
Anthropic announced Tuesday it is developing FlashProve, an open-source system to verify which AI model produced an output. The project aims to provide "provable inference," a cryptographic proof that an answer came from a specific model. It would let users confirm they are using the model they believe they are.
Like cryptographic email signatures, FlashProve would verify a claimed sender. Researchers say it would resist impostors, including the lab that built and runs the model, trying to pass one model off as another. That could prevent a company from secretly swapping in a cheaper model to cut costs or a pricier one to inflate results.
The system could also test third-party claims that an advanced model performed a task. To work, labs must publish a "fingerprint" for each model. The fingerprint can be matched to outputs and verified cryptographically without revealing model weights.
Those weights are the private internal numbers that define behavior, and closed-model makers such as Anthropic and OpenAI keep them secret. Because fingerprints do not expose weights, labs could publish them without surrendering their most valuable asset. The design is lightweight enough for verification on a local device.
Anthropic released a technical specification and working code for system elements on GitHub under an open-source license. Research is ongoing, and there are no plans to deploy FlashProve at Anthropic or any other lab. The announcement calls the work intermediate progress, not a completed research agenda.
The researchers say they are sharing the work so others can examine, adopt, and improve it. If adopted, FlashProve could give users and auditors a stronger way to verify model identity. It could also make AI providers more accountable for the models they actually serve.
