Integrity

Confidentiality is the promise. AI is a tool we keep under custody — inside the perimeter, on the record, and never holding the pen.

Perimeter

AI touching manuscript content only runs inside the platform, on models with no retention and no training, under the policy each journal chooses.

Provenance

Every AI action is recorded: what it did, when, and whether the human accepted or rejected it. Disclosure is generated from the trail, not self-declared.

Authority

Tools produce evidence, checks and questions — never a verdict, score or recommendation. The reviewer signs the report and decides item by item.

What stays forbidden

Generating the review report
Assigning a score or recommendation
Sending the manuscript to any system outside the perimeter
Training models on manuscripts or reports
Operating without a provenance record

Your AI policy, configurable per journal

Editors pick the level. The audit trail is exportable at every level, so a journal that rewrote its AI policy in 2025–26 can show exactly what happened.

Level 0: No AI touches the manuscript at any point.
Level 1: Metadata only: references, conflicts and formal checks.
Level 2: In-perimeter copilot for the reviewer, with full provenance.

Available today: closed perimeter, confidential storage, human-signed reports and a verifiable record. On the roadmap: metadata layer, in-perimeter copilot, exportable provenance trail.

Probatus never sees or influences the editorial decision. Reviewers stay anonymous, conflicts of interest are blocked before matching, and the journal keeps full authority over the outcome.