Grant Thornton’s 2026 AI Impact Survey, published this week, canvassed 950 senior US business leaders and found that 78 percent lack confidence their organization could pass an independent AI governance audit within 90 days. Only 12 percent say their workforce is genuinely AI-ready, with the remaining majority describing readiness in softer terms.
The numbers beneath the headline are more striking. Seventy-three percent of the organizations surveyed are already running autonomous or near-autonomous AI systems in production. Only 20 percent have tested a failure response plan for those systems. Five percent permit high-stakes autonomous decisions without any human review. There is also a significant perception gap within the same organizations: CIOs and CTOs report 47 percent workforce readiness, while COOs at the same companies report 9 percent. That gap suggests AI deployment and AI accountability are being tracked by different people with very different visibility into what is actually running.
AI governance, in practice, means being able to demonstrate that an organization knows what AI systems it has, what decisions they touch, how they perform, and what happens when they fail. An independent audit checks for an AI asset inventory, risk classification by stakes level, documentation of training data and model behavior, human oversight protocols for high-risk decisions, and tested incident response plans. ISO 42001, the first certifiable international standard for AI management systems, published in 2023, is the closest thing to a recognized benchmark for what such an audit would actually measure.
A counterpoint worth noting sits in the US legislative picture. A bill working its way through Illinois, backed by OpenAI, would shield frontier AI developers from civil liability for critical harms caused by their models. The bill targets developers like OpenAI and Anthropic, not the enterprises deploying their products. But the direction it represents matters: as enterprise AI deployment outpaces internal accountability infrastructure, one policy trajectory runs toward reducing the financial consequences of the failures that poor governance makes more likely.
Grant Thornton sells AI governance advisory services, and the survey’s finding that companies need more governance serves that business directly. The numbers are nonetheless consistent with comparable work from IBM, Deloitte, and KPMG. An industry where 73 percent of organizations are running autonomous systems and only 20 percent have tested a failure plan does not require a particular methodological posture to find that gap concerning.
Sources: Grant Thornton 2026 AI Impact Survey (950 US senior leaders, February–March 2026); Illinois SB3444
–
By the Control Plane Editorial Team