McKinsey Trust Survey: RAI at 2.3 — Agentic Controls Still the Lag
McKinsey 2026 AI Trust Maturity Survey (~500 orgs, Dec 2025–Jan 2026): average RAI maturity 2.3 (from 2.0), but only ~30% hit level 3+ on strategy, governance, and agentic controls. Security is top barrier to scaling agents (~2/3); training gaps lead RAI blockers (~60%). Consultant survey — self-reported, not an audit.
Adoption is racing into agents; trust tooling is not. McKinsey’s proprietary cut: average RAI maturity ticked up, but agentic governance is where the middle of the market still fails — security fear as the scale brake.
McKinsey (published March 25; fieldwork Dec 2025–Jan 2026) surveyed ~500 organizations with AI governance responsibility using a five-dimension RAI Maturity Model (strategy, risk management, data/technology, governance, plus agentic AI governance/controls). Average score 2.3 (from 2.0). Only ~30% at level 3+ in strategy, governance, and agentic controls. Asia–Pacific leads; tech/media/telecom and financial services outperform other sectors. Significant RAI investment correlates with higher maturity and material EBIT impact (McKinsey).
| Signal | McKinsey finding |
|---|---|
| Top barrier to scaling agentic AI | Security/risk (~2/3 of respondents) |
| Top cited risks | Inaccuracy; cybersecurity |
| Mitigation vs awareness | Active mitigation lags across most risk categories |
| Incident frequency | ~8% (stable); confidence in response declined |
| RAI implementation barrier | Knowledge/training gaps (~60%) |
| Accountability | Explicit RAI ownership → higher maturity |
RAI increasingly framed as business enabler (value, efficiency, trust), not pure compliance.
Claims vs checks
Figures are McKinsey survey primary — self-reported maturity, not third-party audits. Sector and regional leads are within-sample. EBIT linkage is correlational as presented. Cross-check with other RAI surveys before treating 2.3 as industry absolute.
Limits
- Survey of governance-responsible orgs skews toward those who already care.
- “Level 3+” thresholds are McKinsey’s model, not a standard.
- Agentic-control questions are new — year-over-year comparability is partial.
Sources
- McKinsey: “State of AI trust in 2026: Shifting to the agentic era” (March 25, 2026).
- McKinsey AI Trust Maturity Model and survey methodology (~500 organizations).
- Related McKinsey analysis on agentic AI and responsible practices.