Quantitative third-party AI risk. Not a 200-question form.
Vendor AI risk is the spreadsheet nobody wants to own. ShadowIQ scores vendors quantitatively, monitors them continuously, and produces contract-ready artifacts your Legal team can actually use.
Summary
Third-party AI risk assessment in ShadowIQ combines an automated vendor questionnaire with continuous signals (latency, PII handling, training-data policy, subprocessors, breach history) to produce a quantitative vendor risk score aligned to EU AI Act Article 25, NIST AI RMF, and SOC 2.
The before / after, in one picture.
You've heard this one before.
- Different teams assessing the same AI vendor differently.
- Legal reviewing AI contracts without a shared risk rubric.
- No continuous monitoring — assessment is a point in time, not a relationship.
- Subprocessor changes missed until an incident.
Three moves.
- 1One shared rubric.
Quantitative scoring model reviewed with Big-4 advisory. Customizable weights; defaults aligned to EU AI Act Article 25.
- 2Continuous signals.
Automated probes: latency, TLS configuration, training-data policy, subprocessors, public breach history — updated daily.
- 3Contract-ready artifacts.
Assessment summary, SCC addenda, and DPA clauses generated from the score. Legal edits, doesn't draft from blank.
Numbers, not adjectives.
Asked, answered, sourced.
A weighted model across five pillars: data handling, model security, operational maturity, incident history, and contract posture. Each input is either a verified signal or a questionnaire answer — weights and formulas are visible and customizable.
Yes. We ingest SIG, CAIQ, and CAIQ-Lite. The ShadowIQ AI addendum (30 questions) is designed to plug into your existing TPRM tool (OneTrust, ProcessUnity, SecurityScorecard).
We monitor subprocessor pages, DPAs, and public filings; any change triggers a reassessment and a delta is added to the signed record. Legal is notified in ServiceNow or Jira.
Keep going.
Your 30-minute demo. A signed audit trail by the end of it.
We'll wire ShadowIQ into one live workload, block a prompt injection in real time, and hand you a cryptographic receipt — before the meeting ends.