The incident managing brokers fear
An agent asks ChatGPT Enterprise to “draft a counteroffer update for the buyers and loop in the co-listing agent.” The model pulls language from internal notes — including the seller's willingness to accept less and a prior inspection concern. The draft cc line includes a party who should not see seller motivation. One send away from a fair housing or confidentiality complaint.
Why state CE on AI isn't enough
Many states now expect brokerages to train agents on AI use. Training covers competence — not enforcement at the moment of send. Brokerages still need firm-wide controls when dozens of agents use Copilot daily on CRM, email, and transaction platforms.
- MLS rules on advertising and data display
- Client agency agreements and confidential information duties
- Fair housing sensitivity in generated listing and outreach language
- Property management tenant data in mixed commercial/residential shops
Controls a 200-agent brokerage can adopt
Offer and counteroffer email → broker approval
Any message discussing price, concessions, deadlines, or seller motivation routes to designated broker before send.
MLS + private notes separation
Block AI from merging internal strategy notes into client- or co-broker-facing drafts without review.
Commission and referral communications
Messages about fees or referral agreements require compliance or managing broker approval.
Audit trail for DRE or board inquiries
Log what AI accessed, who approved outbound communication, and what was sent.
Property management note
Firms that combine brokerage with property management face tenant PII and lease terms in the same Microsoft tenant. Apply the same pattern: tenant-facing or owner financial communication requires approval; bulk export of rent rolls or applications is blocked and logged.