Posted on by
Arild Aarnes573Super User 2026 Season 2
Hi,
I'm checking out the "Run a multistage approval" action from the Human in the loop section in an agent flow.
I've created a couple of inputs and put in a small prompt in the first stage. I'm trying to compare two dates and if the dates matches it should reject the request and then move to the next stage. The next stag is a manual stage where a person will review the request.
But the problem is AI approves the first stage even though it's own conclusion is that it should reject? See image below
Any idea why this is happening?
This is just a test of the "Run a multistage approval" action. I could just have compared the dates outside of this action and get it to work but I just wanted to test this new action :-)
This may be due to how the AI approval stage interprets the decision versus the explanation it generates. In your screenshot, the reasoning clearly states the request should be rejected, but the approval outcome is still marked as Approved, which suggests the model is generating the rationale correctly but not mapping it consistently to the approval decision.
Since this feature is still in preview, I'd be cautious about using it for deterministic business rules such as date comparisons. For now, I'd recommend performing the date validation with a standard condition before the approval stage and only routing valid requests into the multistage approval process. If the behavior is reproducible, it would be worth raising a support ticket as it appears inconsistent with the documented intent of the AI stage.
Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.