@adsads
AI safety is not only about model behavior.
It is also about what happens before an irreversible action executes.
Decision-OS V5 Revised (SiriusA2) proposes a confirmation layer for human–AI decision safety:
intent → observe/hold → approve/reject → execute/revoke
It combines:
* trajectory-aware duress routing
* two-step confirmation
* revoke path
* family multisig
* ZK authorization without exposing PII
* non-PII audit evidence
ZK does not read the user.
It proves authorization.
That distinction matters.
Preprint: <https://zenodo.org/records/19828435>
0 reply
1 recast
4 reactions