When an LLM reads a lead record, it doesn't stop at the name and company alone; it reads every field on that record, including revenue, phone number, and home address, unless told not to. Zoho CRM has introduced AI Governance to answer exactly this question: what should an LLM actually be allowed to read from your records?
For every field in a module, you can now choose whether an internal or external LLM can access the data it holds, and if it can, whether the value should be masked rather than shared in full. Some fields hold data that's perfectly fine for a model to use; others hold sensitive details, such as national ID numbers or revenue figures, that you'd rather keep out of an AI's reach entirely, or let it work around without ever exposing the real value.
AI Governance lives in the Fields tab of every module, covering all module types: standard (Org), Custom, and Team modules. Go to Setup, then Modules and Fields, select the relevant module, open the Fields tab, and choose AI Governance.
For both the Internal LLM (Zoho's own AI) and the External LLM (a model you've connected), you get two choices per field:
Use the search box if the field list is long, set the controls to fit your policy, and save. These settings are per module, so they need to be applied separately for each one.
Consider a Leads module where records store a national ID (such as a social security number or local equivalent) alongside the usual company, city, and revenue fields. Sales reps use an AI assistant to summarise leads and draft follow-ups, and want those summaries genuinely useful, so most fields should remain readable. The national ID, however, should never appear in a generated summary, and shouldn't be handed to an outside model at all.
With AI Governance, the everyday fields stay allowed, the ID field is unticked entirely for the external model, and for the internal model it's allowed but with masking applied. Reps still get rich, helpful summaries built from the everyday fields; the ID stays masked in output generated by the internal model, and is withheld from the external model entirely, so it never appears in its output either. You get the benefit of AI without handing it data it has no business revealing.