Enterprise trust in autonomous AI agents slips, VentureBeat survey finds

Enterprises are getting cold feet about letting AI agents ship changes on their own. According to a VentureBeat survey published October 8, 2026, the share of companies allowing or pursuing automated production changes fell from 75% in July to 56% in August.

That is a 19-point slide in a single month.

What the numbers show

The August data comes from 140 respondents. The core question was whether organizations let AI agents make production changes based solely on automated evaluations, with no human approving the final step.

The respondent pool also shifted. Decision-makers made up 53% of participants in August, up from 44% in July.

Among final AI purchasing decision-makers specifically, willingness to allow automated changes dropped from 88% to 61%. That is a 27-point decline in the group that actually signs the contracts.

The August survey also broke down where companies currently stand on low-risk agents:

• 27% already let low-risk agents deploy changes without human validation.
• 20% are engineering toward that capability within the next 12 months.
• 35% have ruled it out indefinitely.

The problem with passing the test

The most uncomfortable number in the survey concerns evaluations themselves. Among organizations that run pre-deployment evaluations, 61% reported at least one case in the past 12 months where an AI feature passed testing and then failed in a way that affected customers.

Earlier monthly surveys from 2026 had already hinted at this tension. Those surveys found only 5% to 13% of respondents expressed full trust in automated evaluations. They also pointed to approximately 49% to 50% of deployments involving false positives that led to failures, meaning tests flagged something as working when it was not.

Against that backdrop, the July figure of 75% looks like the outlier. August may simply be enterprises recalibrating toward what their own testing experience has been telling them.

What this means

It is worth keeping the scale in perspective. The August survey covers 140 respondents, and the composition of the pool changed between months, with more decision-makers in the latest round.

What to watch next is whether the September and October figures hold near 56% or bounce back toward July’s level. Also worth tracking: whether the 20% engineering toward autonomous low-risk deployments actually gets there within their 12-month window, or quietly joins the 35% that have opted out.