Anthropic has released details of FlashProve, an open-source cryptographic system designed to prove which AI model produced a given output, the company said in a blog post published Tuesday that also makes the technical specification and working code publicly available on GitHub.
The project is built to provide what Anthropic's researchers call "provable inference": a mathematical guarantee that an answer really came from the model the user thinks they are using. Like cryptographic email signatures, which let recipients confirm a message's claimed sender, FlashProve is intended to be robust against impostors attempting to disguise one model as another, even when the impostor is the lab that built and runs the model.
The researchers say the system could stop an AI company from secretly swapping in a cheaper model to cut costs, or a more expensive model to inflate its performance results. It could also verify third-party claims that a specific task was done by an advanced model. "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 write. "How can a user verify these claims?"
To work, the system requires AI labs to publish a "fingerprint" of each model they release that can later be matched to specific outputs and verified cryptographically. The fingerprints do not reveal a model's weights, the internal numbers that define how it behaves and that closed-model makers such as Anthropic and OpenAI keep private, meaning labs could publish the fingerprints without giving up their most valuable asset. The system is designed to be lightweight, so verification could be performed on a local device.
Research on the system is ongoing, and there are no plans to deploy it at Anthropic or any other lab. "This research post is meant to share intermediate progress, and not a completed research agenda," the researchers explain. "We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it."
Analysis
Why This Matters
- If adopted, FlashProve could give users a technical way to confirm they are actually getting responses from the model an app or developer claims to be using, rather than a cheaper or different substitute.
- It targets a growing trust problem in AI services: the difficulty of independently checking claims about which model performed a task.
- Because it works without revealing model weights, the system could apply to commercial closed models, not just open ones.
- Anthropic says the research is not complete and there are no deployment plans, so any practical impact depends on other labs taking up and improving the work.
Background
Anthropic is an AI research company and the maker of the Claude family of language models. As AI models are increasingly offered behind APIs, users rarely have a technical way to confirm which model actually generated the response they received, and providers have an incentive to route requests to less expensive or more flattering models.
What to Watch
- Whether other AI labs, including closed-model makers, publish fingerprints for their models, which FlashProve would need to work in practice.
- Whether Anthropic moves the system beyond research; the company says there are no plans to deploy it yet.
- How the open-source community responds, since the announcement explicitly invites others to take the code and improve it.
Checked
12 claims in this story checked against 1 source: 11 supported.
Show each claim
- SupportedThe researchers’ quoted example and question match the source: “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,” followed by “How can a user verify these claims?”TechCrunch: ““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 write. “How can a user verify these claims?””
- SupportedThe researchers’ closing quoted passage matches the source.TechCrunch: ““This research post is meant to share intermediate progress, and not a completed research agenda,” researchers explain in the announcement post. “We are sharing this explicitly with the goal that others look at what we have done, take it, and improve it.””
- SupportedAnthropic is working on FlashProve to verify which AI model produced an output.TechCrunch: “Anthropic is working on an open-source system to verify which AI model produced a given output, according to a blog post published Tuesday.”
- SupportedFlashProve is an open-source system for verifying which model produced an output, described as “provable inference.”TechCrunch: “The system, called FlashProve, is built to provide what researchers call “provable inference” – a cryptographic proof, essentially a mathematical guarantee, that a given output came from a specific model.”
- SupportedThe model fingerprints do not reveal the model’s weights.TechCrunch: “Crucially, the fingerprints don’t reveal a model’s weights, the internal numbers that define how it behaves and which closed-model makers like Anthropic and OpenAI keep private.”
- Partly supportedThe announcement made the technical specification and working code publicly available on GitHub.TechCrunch: “As part of the announcement, researchers released a detailed technical specification and working code for various elements of the system, which is publicly available under an open-source license on GitHub .”
- SupportedFlashProve provides a mathematical guarantee that a given output came from a specific model.TechCrunch: “The system, called FlashProve, is built to provide what researchers call “provable inference” – a cryptographic proof, essentially a mathematical guarantee, that a given output came from a specific model.”
- SupportedFlashProve is intended to be robust against impostors disguising one model as another, including the lab that built and runs the model.TechCrunch: “It’s meant to let users confirm that the answer they received really came from the model they think they’re using. Similar to cryptographic email signatures that let recipients confirm a message really came from its claimed sender, FlashProve would be robust against impostors attempting to disguise one model as another, even when the impostor is the lab that built and runs the model.”
- SupportedThe researchers say the system could prevent an AI company from secretly swapping in a cheaper model to cut costs or a more expensive one to inflate performance results.TechCrunch: “In particular, the researchers note that the system could be used to prevent an AI company from secretly swapping in a cheaper model to cut costs, or a more expensive model to inflate its performance results.”
- SupportedFlashProve could verify third-party claims that a specific task was done by an advanced model.TechCrunch: “FlashProve could also be used to verify claims by third parties that a specific task was done by an advanced model.”
- SupportedThe system requires AI labs to publish a fingerprint for each model they release, which can later be matched to specific outputs and verified cryptographically.TechCrunch: “To work, the system requires AI labs to publish a “fingerprint” of each model they release that can later be matched to specific outputs and verified cryptographically.”
- SupportedThere are no plans to deploy FlashProve at Anthropic or any other lab.TechCrunch: “Research in the system is ongoing, and there are no plans to deploy it at Anthropic or any other lab.”
