Anthropic Open-Sources FlashProve to Cryptographically Verify Which AI Model Produced a Response
Anthropic has open-sourced FlashProve, a prototype system that uses cryptographic proofs to verify which AI model generated a given output. The system requires labs to publish model "fingerprints" that do not reveal private weights, allowing verification to work with commercial models and to be performed locally. Anthropic released a technical specification and code on GitHub, saying research is ongoing with no plans to deploy it yet. In one test on Meta's Llama 4 Maverick, a single Nvidia H100 GPU produced proofs at 110 tokens per second, each about 19MB, which a regular CPU checks in about 17 seconds.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Anthropic Open-Sources FlashProve Provable Inference Prototype for MoE Models
Anthropic has open-sourced FlashProve, described in the post as its provable inference prototype for Mixture-of-Experts (MoE) models. The announcement was shared via an X post linking to a securitytc.com item. The post frames FlashProve as a prototype tool for provable inference in MoE model settings, indicating a release of code or technical materials by Anthropic. No further technical details, performance figures, licensing terms, or independent evaluations are included in the source text. The item is a brief release announcement rather than a full technical disclosure.
Read sourceAnthropic Develops FlashProve to Cryptographically Verify AI Model Identity
Anthropic announced Tuesday that it is developing an open source system called FlashProve to verify which AI model produced a given output. The system, described as providing "provable inference," uses cryptographic proofs to let users confirm that a response genuinely came from the model it claims to be from, similar to cryptographic email signatures. Researchers say it could prevent AI companies from secretly swapping in cheaper or more expensive models, and could verify third-party claims that a task was performed by a specific advanced model. FlashProve requires labs to publish a "fingerprint" of each model that can be matched to outputs without revealing the model's private weights, allowing it to work with commercial models. The system is designed to be lightweight enough for local verification. Anthropic released a technical specification and open source code on GitHub, but said the research is ongoing with no plans to deploy it yet, and invited others to improve upon the work.
Anthropic Develops FlashProve to Cryptographically Verify AI Model Identity
Anthropic announced Tuesday that it is developing an open-source system called FlashProve to verify which AI model produced a given output. The system provides "provable inference" — a cryptographic proof that an output came from a specific model — similar to cryptographic email signatures. It is designed to resist impostors, including the lab that built the model, and could prevent companies from secretly swapping in cheaper or more expensive models. FlashProve could also verify third-party claims that a task was performed by an advanced model. The system requires labs to publish a "fingerprint" of each model that does not reveal its weights, allowing it to work with commercial models, and is lightweight enough for local verification. Anthropic released a technical specification and open-source code on GitHub, but said the research is ongoing with no plans to deploy it.
Read sourceShow 2 older updatesHide older updates
Anthropic open-sources FlashProve to verify which AI model wrote a reply
Anthropic has open-sourced FlashProve, a system that lets a server mathematically prove which AI model generated a given reply, with verification checkable on an ordinary CPU. The tool addresses cases such as resellers quietly serving a cheaper model than the one a customer paid for. Anthropic tested FlashProve on Meta's Llama 4 Maverick, a 400B-parameter open model, where a single Nvidia H100 GPU produced proofs at 110 tokens per second, each about 19MB. A regular CPU checks each proof in about 17 seconds, compared with VerInf, the only earlier system known to prove this model, which produced 93.6GB proofs that took almost a day to check. Claude built the system, and because its code is hard to audit, the security guarantees are mathematically proved in Lean 4, a proof-checking language.
Anthropic Releases FlashProve for Verifying LLM Response Provenance
Anthropic announced new research introducing FlashProve, a cryptographic technique that lets developers verify whether a given response came from a given AI model. Anthropic says FlashProve is substantially more efficient than prior open-source methods for this purpose. The proofs have applications such as making it harder for attackers to hijack a chat and enabling safety evaluators to verify that the model they are testing is the same one being served to customers. Anthropic stated that further progress in provable inference will likely be an important part of improving trust in AI deployment, and it is open-sourcing FlashProve's code. The company credited collaborators Alessandro Chiesa, Giacomo Fenzi, G. W. Berg, Miranda Christ, Rahul Dalal, and Ben Mayer.
Read source