aifollow.news
← TrendingLatest development Nov 04 04:15
STORY IN FOCUS

Anthropic is developing FlashProve to verify which model produced an AI output

FlashProve aims to let users verify whether an answer came from a specified model. Anthropic researchers have published technical specifications and runnable open-source code. The model “fingerprints” needed for verification do not reveal model weights, and users can verify outputs on local devices. Research is ongoing, and Anthropic has no plans to deploy the system yet.

1 developments4 reports3 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

Anthropic open-sources FlashProve to verify which AI model generated a response

Anthropic has introduced and open-sourced FlashProve, which lets developers verify whether a response came from a specified model. One proposed use is checking whether the model evaluated for safety is the one served to customers. The source reports that a regular CPU checked each proof in about 17 seconds in a Llama 4 Maverick test; that result does not establish performance for other models.

View source reports · 3 sources · 4 reports
All reports · 4

By original publication time, with each report's bookmarks and feedback preserved.

IT之家 科技新闻 Machine translation

Anthropic is developing FlashProve to verify which model produced an AI output

FlashProve aims to let users verify whether an answer came from a specified model. Anthropic researchers have published technical specifications and runnable open-source code. The model “fingerprints” needed for verification do not reveal model weights, and users can verify outputs on local devices. Research is ongoing, and Anthropic has no plans to deploy the system yet.

TechCrunch AI Machine translation

Anthropic develops cryptographic system to verify AI model identity

Anthropic is developing FlashProve to let users verify whether an answer came from a specified model. The system requires model providers to publish fingerprints that can be matched to outputs without revealing model weights. Researchers have released a technical specification and working code for parts of the system, but there are no plans to deploy it.

Rohan Paul (@rohanpaul_ai) Machine translation

Anthropic open-sources FlashProve to verify which AI model produced a response

Anthropic has introduced FlashProve in new research and open-sourced its code, allowing developers to verify whether a response came from a specified model. It could help safety evaluators check that the model they test is the one served to customers. Anthropic says it is substantially more efficient than earlier open-source methods.

Rohan Paul (@rohanpaul_ai) Machine translation

Anthropic open-sources FlashProve to verify which AI model generated a response

Anthropic has introduced and open-sourced FlashProve, which lets developers verify whether a response came from a specified model. One proposed use is checking whether the model evaluated for safety is the one served to customers. The source reports that a regular CPU checked each proof in about 17 seconds in a Llama 4 Maverick test; that result does not establish performance for other models.