IETF 127 hackathon shows independent verifiers disagreeing on identical evidence data
At the IETF 127 Hackathon in San Francisco in November 2026, several teams building verifiable records of AI agent actions had outside developers write verifiers from their documents alone. Each implementation passed its own tests, but the verifiers diverged when run against one another. One project's two reference verifiers gave different serializations of the same number and reached conflicting ALLOW and DENY verdicts. A separate comparison of four JSON canonicalization implementations produced differing bytes on a small share of test cases, mainly involving very large integers, byte order marks, duplicate members, and deep nesting. Operators comparing receipts from two transparency logs also found the same statement yielded different leaf hashes, because the specifications do not fix the entry serialization. The authors say these results point to the need for explicit rules, such as refusing ambiguous inputs, to keep verification consistent across implementations.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in