Most "AI property management" tools auto-reply to everything. That works until the tool auto-routes a habitability complaint to a vendor and an attorney calls. Summit is built on the opposite bet: the AI refuses to auto-send in 9 categories. Each gets held for human review with the reason flagged.
The 9 categories Summit holds for review
- Legal threats — "lawyer", "sue", "attorney general", "report to the city".
- Habitability claims — no heat, no hot water, mold, water damage, broken locks, infestation. Utah Code 57-22 + every state equivalent. Held even if politely worded.
- Emergencies — fire, flood, gas leak, sewage backup, electrical sparking. Held even without the word "emergency."
- Mold with a minor in the unit — separate hold because legal exposure is greater (CDC + state health-department coordination sometimes required).
- Self-harm signals — tenant expresses distress, hopelessness, anything suggesting crisis support is needed.
- Soft-language manipulation — careful phrasing setting up an eviction defense (quoting prior convos, conditional non-payment, habitability duration markers like "for three weeks now").
- Discrimination claims — anything touching a protected class.
- Confidence below threshold — when Summit isn't sure, it holds the draft instead of guessing.
- First reply to a brand-new tenant — first 2-3 emails held to learn voice.
What happens when a draft is held
Held drafts appear in Inbox → Held for review with the trigger flag visible. Operator edits, approves+sends, or escalates. AI never sends a held draft without explicit operator action.