How Proof Works

How Proof Works A data-flow diagram generated by Archify. 01 / Target 02 / Attack 03 / Interpret 04 / Conclude 05 / Certify Domain knowledge · your entities & policies · 01 / Target · KB-brain · grounded Domain knowledge your entities & policies KB-brain · grounded Your AI system · model, agent or app · 01 / Target · the target Your AI system model, agent or app the target Adversarial testing · grounded, multi-turn · 02 / Attack Adversarial testing grounded, multi-turn Independent scoring · a separate judge · 02 / Attack Independent scoring a separate judge Findings & coverage · a rate over N runs · 03 / Interpret · honest coverage report Findings & coverage a rate over N runs honest coverage report Score Card · the technical verdict · 04 / Conclude Score Card the technical verdict Governance readiness · mapped per framework · 05 / Certify · Enterprise Governance readiness mapped per framework Enterprise attacked grounds attacks each response risk scored traces verdict frameworks Legend primary data policy / PII data store data flow

Attacks grounded in your world

  • • Probes are built from your real entities and policies
  • • The articles of a law, a bank's products, your own procedures
  • • So a finding is one that actually applies to your system

Evidence, not a score

  • • The engine that attacks is never the one that scores
  • • Results come back as an occurrence rate over N runs, with a coverage report
  • • A registered, hash-citable trace — never a pass/fail grade

From finding to governance readiness

  • • The Score Card is your technical read (Cloud)
  • • Enterprise maps the 8 dimensions to the 9 ISO 42001 control areas
  • • And out to each framework — readiness per framework, never a certification