Science & Tech · AI · 4 hrs ago
Anthropic develops open-source system to verify AI model outputs
Anthropic is developing FlashProve, an open-source system designed to verify which AI model produced a particular answer.
The system would use cryptographic proof to let users check that an output came from the model they were told was being used.
This could help guard against companies secretly switching models, for example to cut costs or make performance results look better.
To use it, AI labs would publish a fingerprint for each model that can be matched to outputs without revealing the model’s private internal numbers.
Anthropic says the system is designed for verification on a local device, and it has released technical details and working code publicly.
Research is ongoing, with no deployment planned, and the team is inviting others to test and improve the system.
Anthropic is developing FlashProve, an open-source system to verify which AI model produced a given output.
The system aims to provide a cryptographic proof that an output came from a specific model.
It could help users detect when a model has been secretly replaced or verify claims that a particular model performed a task.
Anthropic researchers released a technical specification and working code under an open-source license.
Research is ongoing, and there are no plans to deploy FlashProve at Anthropic or another lab.
- Who
- Anthropic researchers are developing FlashProve.
- What
- An open-source system intended to cryptographically verify which AI model produced an output.
- When
- The system was announced on Tuesday, November 3, 2026.
- Where
- Not stated
- Why
- To let users verify model identity and detect possible substitutions or unsupported claims about which model performed a task.
This story does not have two clearly opposing sides.
Anthropic researchers announced FlashProve in a blog post.
Researchers released a technical specification and working code under an open-source license.
Research is ongoing, with no plans to deploy the system at Anthropic or another lab.
- System name
- FlashProve
- Developer
- Anthropic researchers
- Purpose
- Cryptographically verify which model produced an output
- Release
- Technical specification and working code made publicly available under an open-source license
- Deployment plans
- None at Anthropic or another lab, according to the researchers
What people are saying
This research post is meant to share intermediate progress, and not a completed research agenda
We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it.