ResearchBeyond tamper-evident receipts: why AI evidence needs an independent witness
A hash chain proves a set of records is internally consistent; it does not prove that set of records is the one that existed yesterday. That gap between the two claims is where the interesting attack lives, and better hashing does not close it.
ResearchWe read the code behind a dozen self-improving AI agents
We cloned, installed and in several cases live-tested twelve self-improving agent frameworks. Nine of ten could not prove their own history had not been altered. This is the pattern that came out the other side.
ResearchThe two questions that actually decide if an autonomous agent is safe
Audit logs answer 'can you prove what the agent did'. The question a business actually asks is simpler: will it do the wrong thing, and will I find out before it matters. We looked hard at both, and one is unsolved everywhere, including here.