I’m experiencing an issue with my custom app in Amazon Quick. The app is returning the following error for ANY input — including simple messages like “hi”:
Error: “This request could not be processed due to content policy violations. Please try again with a different prompt.”
Details:
The app is designed for investigation case handling (NI Abuse Type Detector)
The persona instructions contain Standard Operating Procedures (SOPs) for abuse investigation workflows
The SOP content includes terminology related to: PDP tampering, NasD (Not as Described), content abuse, image abuse, miscat, duplicate ASINs, etc.
The error occurs regardless of what the user types — even “hi” triggers it
Other apps I created work fine — the issue is isolated to this specific app
I have NOT changed anything recently — it was working before
What I’ve tried:
Sending different types of messages (all blocked)
Testing other apps (they work normally)
Questions:
Is the content policy filter being triggered by the SOP content in the persona instructions/reference documents?
If yes — is there a way to whitelist or exempt internal investigation SOPs from the content filter?
Can you check what specifically is triggering the violation in my app’s configuration?
Is there a recommended approach for apps that need to reference sensitive operational content (abuse types, enforcement actions, policy violations) without triggering the content filter?
Expected behavior: The app should process user queries and reference the SOP content to assist with investigation workflows.
App purpose: Assist investigators in following correct procedures for different abuse types in the CDT/NI workflow.
Please advise on the best path forward. I need this content for my daily work and cannot remove it.
Yes, the content filter is almost certainly being triggered by the SOP content in your persona instructions/reference documents. The way it works is that the filter evaluates the entire prompt context on every request, not just what the user typed. So your SOP terminology (abuse types, enforcement actions, etc.) gets evaluated alongside “hi,” and the combined payload is what’s getting flagged. That’s why every input fails even though you haven’t changed anything and other apps work fine. The Limitations docs acknowledge this can happen (“Prompts occasionally trigger security guardrails as false positives”), but in your case it’s not occasional since the triggering content is permanently baked into the prompt context.
Unfortunately there’s no whitelist or exemption mechanism available in the UI today. From my experience, the best workaround is moving your SOP content out of persona instructions/reference documents and into a Space instead. Spaces work differently because they use retrieval-based search, so content only gets pulled into the prompt when a user’s query is relevant to it rather than being shoved into every single request. I do not think it will guarantee zero filter hits (if a specific passage gets retrieved that contains the sensitive terms, it could still trip), but ideally it should stop the blanket blocking you’re seeing now. If that’s not feasible, another idea I can think of is try abstracting the sensitive terminology by replacing the literal abuse-type labels with internal codes.
I’d also recommend opening an AWS Support case. The filter thresholds aren’t user-configurable, but your use case (referencing operational SOPs for investigation workflows) is a legitimate business need that AWS has written about getting caught by safety mechanisms. A support case at minimum creates visibility and might surface options specific to your situation.
Just checking back in since we haven’t heard from you in a bit. I wanted to see if the guidance shared earlier helped resolve your question, or if you found a solution in the meantime. Moreover, if you submitted a support ticket and it uncovered any findings, please feel free to share them!
If you still have any additional questions related to your initial post, feel free to share them. Otherwise, any update you’re able to provide within the next 3 business days would be helpful for the community.
Since I haven’t received any further updates from you, I’ll treat this inquiry as complete at this time. Please feel free to create a new post if you have more questions.