Compliance Glossary for Confluence

Can AI draft glossary terms?

Yes. In production v6, a user with term-management access can generate editable candidate previews from selected undefined-term findings. Forge LLM uses bounded, redacted evidence from current Confluence page revisions. A preview becomes a normal Draft only after human review, scope selection, and explicit creation. AI cannot approve, publish, overwrite a term, or change a Confluence page.

Last verified against production v6 · 26 September 2026

At a glance

ControlProduction v6 behaviorBoundary
AccessActive license and term-manager gateIf no RBAC managers exist, management access is open
RequestUp to 10 selected candidates and three page IDs eachOnly undefined-term findings qualify
EvidenceTwo matching sentences per page, up to 500 characters eachSecret-like assignments are redacted
OutputProposed term, aliases, optional expansion, and draft definitionOutput can be incomplete or wrong
PersistenceCandidate wording, page references, versions, model metadata, and audit metadataNo excerpts, page bodies, prompts, or raw model responses
RetentionUncreated previews expire after 14 daysA daily job removes expired previews
ApprovalCreated terms enter the normal Draft stateAnother eligible person must approve them

How it works

  1. In Compliance Scanner, choose a space and select Scan Space.
  2. Under Top Undefined Terms, select Select for AI draft.
  3. Select Generate draft candidates.
  4. Review the live evidence excerpts and page versions.
  5. Edit Proposed term, Aliases, Draft definition, and any contextual expansion.
  6. Select Create as a draft and choose Global or Space-Scoped.
  7. Select Create selected drafts.
  8. In Terms, use Submit or Submit for Review. Another eligible person completes approval.

Example

A scan finds an undefined acronym on several pages. A term manager generates a candidate and checks the current excerpts. They correct the definition and choose a scope. The app creates a normal Draft. Another eligible person reviews the edited term before approval.

What this does not mean

Draft generation is separate from scan-time AI classification. During a scan, Forge LLM may classify approved-term usage as aligned, misaligned, or unclear.

That classifier receives approved definitions and bounded usage excerpts. It stores verdict metadata, not excerpts. If unavailable, usage remains unchecked and scanning continues.

Scan-time classification does not create term candidates. Draft generation runs only after a term manager selects undefined findings and requests candidates.

AI output is assistance, not authority. It does not provide legal advice, decide regulatory applicability, approve terminology, or change Confluence content.

Related questions

Next step

Generate a candidate, verify its evidence and wording, then send the resulting Draft through human approval.

View on Atlassian Marketplace
Verification basis: production v6 code, tests, manifest, and operations guidance. Verified 26 September 2026.