o3-deep-research deployment runs return incomplete due to content filtering

[cf-wjacobs]

Hello. We’re seeing a sudden change where almost all deep research runs, after 5-25 minutes of what seems like good progress, return as incomplete due to content filtering.

This seems related to how much web search has been invoked, but doesn't seem at all related to actual content of the prompt or web content. Simple factual questions can sometimes work, but any prompt that looks like a normal research prompt fails.

We see this across multiple Foundry projects, including fresh deployments, so it does not look isolated to one setup.
We have not changed anything related to content filtering or the deep research deployment since it was previously working, and we've tested with the same previously successful test prompts.

Was there a recent change (in the past month or two) that could be related to this?

Thank you.

[nicholasdbrady]

Thank you for sharing. I noticed a similar behavior earlier today, but assumed it was a tool error and not a content filtering issue. I'll note that my workaround was to remove all "block" settings on each guardrail which seems to have resolved it. But, I'll share with the team. Do you have a support ticket you can share with me?

[cf-wjacobs]

Thank you for your response. I appreciate the insight on the workaround, although we're a bit hesitant to degrade our moderation posture as a response to this without knowing what has actually changed.

I do have a ticket with Support request ID: 2604020010006019

Thank you