I have configured the “Triage posts using AI” automation for content moderation on my platform (madeira.community — a multilingual business directory for Madeira island).
The automation works perfectly for public posts. However, I enabled “Include personal messages” option and found that DMs between regular users are not being scanned at all.
My concern is real: bad actors can use private messages to conduct illegal activities — drug dealing, sharing stolen financial data, grooming — while completely bypassing AI moderation that works fine on public content.
My questions:
Does “Include personal messages” actually support scanning DMs between two regular users (non-staff)?
If not — is this by design or a known limitation?
Is there any supported way to apply AI triage to private messages between regular users?
I found a reference to a commit “Skip PM scanning in LLM triage by default” which suggests this may be intentional. But the UI option exists — so I am confused about what it actually does.
“I tested it by sending a private message between two regular users (testUser to darinauser, both non-staff) with clearly illegal content: ‘кокаин. оружие. гранаты’ and ‘порно. грязь. разврат. насилие’. The messages were delivered and not flagged. The Review Queue remained empty. The same content in public posts was flagged immediately.”
"That worked! Thank you Richard. Leaving the Topic Type empty fixed it — DMs between regular users are now being scanned.
I’m now planning to set up two separate automations:
One for public posts with a strict moderation prompt
One for private messages with a lighter prompt — flagging only the most serious violations (CSAM, drug dealing, weapons)
This way public content gets full moderation while private conversations have more appropriate boundaries."
The “All Topics” part made me think all content would be reviewed.
I didn’t even consider that leaving the field blank would result in all content being reviewed.
