Why does AI make up names of people who don't exist at my company?
Because it has no people-graph to check against. bRRAIn's POPE layer stores every real person, their role, their email, and their active projects. The Handler checks generated names against the graph and flags fabrications before they reach the user. The model literally cannot invent a colleague.
Fabricated colleagues are a specific failure mode
LLMs invent people's names because they pattern-match on "a manager at your company is called [name-shaped string]" and produce a plausible name. Without a canonical list of real people, the model has no way to refuse. It generates "Sarah from Finance" even when no Sarah exists. This is worse than a generic hallucination because it lands in contexts where the user might act on it — scheduling a meeting, drafting a message, routing a ticket — and the mistake propagates through other systems before anyone catches it.
POPE People nodes are the real org chart
The POPE graph in bRRAIn stores every real person as a Person node with role, email, team, active projects, and status. Ingesting your HR directory, Slack membership, or identity provider populates it automatically. Queries that mention a person resolve against this canonical list. The Memory Engine returns real Person nodes with provenance, not fabricated names. Your org chart stops living in HRIS screenshots and starts being the backbone your AI actually reasons over.
The Handler catches inventions at exit
Even with a grounded retrieval, the model might still slip in a generated name. bRRAIn's Handler runs a final check: any person-shaped token in the output must resolve to a Person node in the graph, or it gets flagged. Unresolved names either get replaced with "someone on your team" or trigger a warning to the user. The Security Policy Engine adds an extra layer — even real names can be suppressed if the querying user isn't cleared to know them. Fabricated colleagues and leaked colleagues both die in the same pass.
Closing the loop with live identity
bRRAIn keeps the Person graph fresh through integrations with Okta, Azure AD, Google Workspace, and HRIS connectors. When someone joins, their node appears; when they leave, their node is tombstoned. The AI learns of org changes automatically and stops referring to ex-employees. Combined with the Handler check, the result is an assistant that cannot invent a colleague and cannot forget a real one. That's the minimum bar for trust in any AI that participates in actual team workflows.
Relevant bRRAIn products and services
- POPE graph / Memory Engine — canonical Person nodes sourced from identity systems.
- Handler — rejects any person-shaped output that doesn't resolve to a real node.
- Security Policy Engine — also suppresses real names the user isn't cleared to know.
- Integrations index — Okta, Azure AD, Google Workspace, HRIS connectors keep the graph live.
- Ontology Viewer — inspect the People graph directly to confirm coverage.