Brief
Preprint Proposes a One-Time Permit for AI-Assisted Claims, Tested Only by Its Authors
A preprint proposes a machine-readable procedure for making AI-assisted claims independently challengeable, and reports its checks as passing. Every reported test is described as in-house or same-host, so what is on offer is internal coherence, not outside confirmation.
The paper proposes Publication Authority: a single-use, non-transferable permit that one publication step must consume. Six obligations cover evidence, runs and artifacts, measurement disclosure, authorization, surface correspondence and lifecycle continuity. A record yields the permit only if all six pass fresh and complete, and none compensates for another.
The testing is internal to the authors. Ten models explored 110,764 safe reachable states, and 76 unsafe configurations produced the expected violation or observer countermodel. A later in-house, instance-blind test matched 183 scored expectations, and same-host package execution reproduced 240 archived observations. The authors say their results support internal coherence, bounded safety, fault sensitivity and limited constructibility.
Our reading
For readers who see AI-assisted claims presented as verified, this is a proposal about what would have to be checked before a claim is allowed to count as published, and it is candid that passing its own correspondence check is not the same as authorization. The reported safety and fault-sensitivity findings come from models and archived runs rather than from an outside party, so the practical va…
What to do or watch
Watch for independent reproduction of the 110,764-state exploration and the scored case classes, and for any move from a candidate specification toward a standing standard; neither external replication nor standards status is reported here.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by arXiv
- The paper develops Publication Authority as an exact-state, non-transferable, single-use publication capability and instantiates it in PAC-2026, described as a machine-readable AIJIM Protocol candidate.
- Six obligations govern evidence, runs and artifacts, measurement disclosure, authorization, surface correspondence, and lifecycle continuity, and none can compensate for another.
- Ten models explored 110,764 safe reachable states, and 76 unsafe configurations produced the expected violation or observer countermodel.
- A later in-house, instance-blind test of known case classes matched all 183 scored expectations, and same-host package execution reproduced its 240 archived observations.
- The authors state that a reader surface passing its correspondence check cannot authorize publication unless the accepted record admits that surface.
Sources
- arXivText stored 17 September 2026
How this story was checked. Written from the 1 page listed above, stored 17 September 2026; claims checked against that stored text on 17 September 2026.
What that means
- 5 of 5 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.