Survey: Only 17% Trust AI Without Oversight

For many teams, AI is now part of everyday work, but most users say it is not dependable without humans in the loop. According to the Connext Global 2026 AI Oversight Survey Report, only 17% of U.S. adults who use AI at work say AI can run on its own with minimal human involvement, while 70% say reliability comes from AI plus light review (35%) or AI paired with dedicated human oversight (35%) – a signal that agentic autonomy still isn’t the default most workplaces trust.

Connext Global’s survey of 1,000 U.S. adults ages 18 and above who use AI in their day-to-day work was conducted in January 2026 via the third-party platform Pollfish. It found that reliability is defined less by automation and more by the human systems around it. The results point to a growing gap between expectations for faster output and the oversight required to keep quality, context and accountability intact across real-world workflows.

Key findings from the report include:

• More than two-thirds (70%) say reliability comes from a human safety net, citing AI plus light review (35%) or AI plus dedicated oversight (35%), compared with 17% who say AI can run on its own.
• Nearly two-thirds (64%) expect the need for human review or checking to increase, with 26% saying significantly and 38% saying somewhat.
• Ongoing monitoring is the norm, with more than four in five (82%) saying AI needs attention almost every time (28%) or sometimes (54%), while just 4% say it can usually run without much attention.
• The “after” work is built in, with nearly all respondents (96%) saying they do follow-up work, while just 4% say they rarely do, most often involving editing or fixing (42%) and review or approval (34%).
• Only 37% say AI is right without fixes most of the time, while nearly two in three (63%) say it is right only sometimes or less (45% sometimes, 16% rarely, 2% almost never).
• When AI output needs fixing, nearly half (46%) say it takes about the same time as doing it manually and 11% say it takes more time.
• Context loss leads to breakdowns, with 42% saying AI left out important details or context, 32% saying it caused extra work to fix or redo and 31% saying it sounded confident but was wrong.
• Customer impact is not rare. About one in five (19%) say AI made a customer situation worse.
• Negative impact is often firsthand, with 60% saying they have personally been involved in AI negatively affecting outcomes, including 18% frustration or complaints and 11% lost revenue or churn.

The findings suggest that as AI adoption expands, autonomy is not the standard most users trust. Instead, reliability is increasingly defined by the workflow around the tool, including who reviews output, how exceptions are handled and when work escalates. This is especially true in higher-stakes, customer-facing moments.

The survey also highlights a practical constraint on productivity. While AI can accelerate drafts, many users report that editing, validation and rework are now built into the process. When errors involve missing context or confident mistakes, teams may spend as much time correcting output as they would completing the work without AI. That makes quality assurance a key determinant of ROI.

“AI is a powerful accelerator, but most teams are still doing the hard work of making its output accurate and complete enough for real-world use,” said Tim Mobley, president and CEO of Connext Global. “The real opportunity isn’t just adopting AI. It’s building the oversight habits that keep quality high as speed increases, so early experiments become lasting advantages.”

As organizations push AI into higher-stakes workflows, the report indicates that competitive advantage may hinge less on deploying new tools and more on building repeatable review, escalation and accountability systems that prevent mistakes from reaching customers.

For the full results and additional insights, view the Connext Global 2026 AI Oversight Survey Report.