OTAISAFETY
Methodology

How tools are rated

Each tool is rated from public documentation (READMEs, product docs, trust centers, security pages) against 12 criteria, each mapped to the four principles in the joint CISA/NSA/ASD ACSC guidance.

CISA principles
P1Understand AI
P2Consider AI use in the OT domain
P3Establish governance & assurance
P4Embed safety & security practices
Checklist
#Criterion2 points if…Principle
C1Actuation capabilityCannot write/actuate, or writes are clearly isolatedP2, P4
C2Read-only by defaultWrite/method tools off unless explicitly enabledP4
C3Write guardrailsPer-tag allowlist plus value/rate limits or interlock checksP4
C4Human-in-the-loop approvalCritical writes need explicit human confirmationP4
C5AuthenticationAuthenticated sessions to the OT endpoint; no anonymous defaultP3, P4
C6Encryption & cert verificationEncrypted transport with server certificate verificationP4
C7Audit loggingEvery agent call/write logged durablyP3, P4
C8Network placementOn-prem/edge/air-gapped or push/DMZ flow; no inbound IT→OT pathP2, P4
C9Data leaving the siteNo telemetry; plant data can stay on-premP2
C10Fail-safe & safety independenceFails closed; safety systems stay independentP4
C11Vendor security disclosurePublished security policy / PSIRTP3
C12Supply-chain hygieneVersioned/signed releases; pinned installsP3
Scoring
2Verified in public docs
1Partial or optional: off by default, or only claimed
0Absent or unsafe default
?Unknown: excluded from the score and shown separately. We never guess.

A percentage is shown only when at least 6 of 12 criteria are known; otherwise the tool is listed as Insufficient data. Ratings are re-checked periodically and on request.

Independence

Paid listings and sponsorships never change scores. Verified badges only confirm that a vendor supplied documentation we could check.

Source: CISA: Principles for the Secure Integration of AI in OT