What the research proposes
Anthropic published FlashProve on 2 November: a research prototype using cryptographic proofs to link a response to previously committed model weights without revealing them. The publication describes limitations and extra computation.
What changes when reviewing an answer
Computational provenance and content accuracy are different checks. Even if the model can be verified, a business still needs to check facts, permissions and results. This is neither an iabot feature announcement nor a guarantee against errors.
Sources and documentation
Provider announcements and results are attributed to their sources. Use cases are editorial proposals, not our own tests or confirmation of model availability in iabot.