Half of Enterprises Cannot Trust Their AI Agents' Decisions, Survey Finds
What happened
The Scaling AI agents with trustworthy data report, published by MIT Technology Review Insights in partnership with Google Cloud, surveyed 300 data and technology executives globally about the state of agentic AI deployment in their organizations. The findings reveal a stark split: approximately half of respondents said they trust the decisions their AI agents make, while organizations classified as high-performing 'data leaders' reported 100% trust in agent outputs. The report identifies weak data infrastructure as the primary barrier to effective agentic deployment, and names data and AI governance enriched with business context as a top organizational priority. Auditability and agent-decision trust frameworks are highlighted as areas where most enterprises remain underprepared. The findings reinforce a pattern visible across recent enterprise AI adoption research: organizations are deploying agents faster than they are building the governance foundations those agents require to be defensible.
Why it matters
- ·If roughly half of enterprises cannot trust their agents' outputs, they almost certainly cannot audit those outputs to a standard that satisfies regulators or affected parties under emerging automated decision-making frameworks. This is not a technology shortcoming, it is a control gap that compliance teams own.
- ·The direct correlation the report draws between data governance maturity and agent decision trust means that organizations without documented data lineage and quality controls (DGC-001, DGC-004) are operating agentic systems on foundations they cannot validate, creating material exposure when those decisions are challenged.
- ·As agentic AI deployment scales, the absence of foundational data readiness will compound audit log integrity and human oversight failures. Compliance teams that have not yet mapped data infrastructure quality to their agent governance programs are likely to find that gap surfaced by regulators before they find it themselves.
Governance controls affected
What to do now
- ☐Map your agentic AI deployments against your current data lineage documentation to identify which agents are operating on data inputs that have not been validated for quality, completeness, or provenance.
- ☐Assess whether your agent audit logs capture sufficient context about the data sources used in each decision to support after-the-fact review by compliance or legal teams.
- ☐Conduct an agentic deployment readiness review (AGT-016) that explicitly includes data infrastructure readiness as a gate criterion, not just model or API readiness.
- ☐Survey internal stakeholders who rely on agent outputs to determine whether trust levels in your organization resemble the survey's low-performing cohort, and document findings as part of your AI risk register.
- ☐Identify any automated decisions made by agents in regulated contexts (credit, employment, healthcare, procurement) and confirm that the underlying data governance controls are sufficient to support explanation or appeal obligations.
What to watch next
Regulators across multiple jurisdictions are moving toward explicit requirements for explainability and auditability of automated decisions, and data quality will increasingly be treated as a precondition for compliance rather than a best practice. The UN Independent International Scientific Panel on AI: Preliminary Report on Agentic AI Governance and emerging automated decision-making rules in the US and EU are likely to sharpen expectations around data readiness specifically. Compliance teams should also watch for follow-on enforcement actions in sectors where agentic decisions already carry legal weight, as the gap this survey documents will be difficult to defend in the absence of structured data governance controls.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.