How AI work in Syllabium is reviewed, checked and recorded.
Can an AI Agent Publish Content?
Syllabium’s authoring assistant does not publish content: it writes drafts and suggestions, and a person publishes through the publishing gate. On content flagged safety-critical, an AI agent’s work always waits for a person, and no setting changes that.
A publish path of the assistant’s own is planned. Under Syllabium’s graded authority, an AI agent’s action above ordinary publishing always waits for a person.
Who Can Approve an AI Change?
Only a person can approve a change in Syllabium, because an AI agent never approves. For content flagged safety-critical, the approver is a second person, not the one who submitted the change.
Where a capability is set to Admin review, a different administrator decides. Loosening a safety-critical setting also waits for a different administrator.
What Does a Reviewer See?
A reviewer in Syllabium sees the held change compared word by word, then approves it or rejects it with a reason.
Staff see the mark “(AI)” and the agent’s role beside AI work, and content the assistant changed is marked.
Is the Author Told Before Publishing?
Yes. Syllabium tells the author, before the click, whether the publishing gate publishes, holds or refuses the change.
How Long Can a Change Wait?
In Syllabium, each kind of change can carry a reviewer chain, a time limit and an escalation, set in its editorial workflow.
How Is Each AI Action Recorded?
Syllabium records every attempted AI action with the check that decided it, and every model call with its model, tokens and cost. These records are tamper-evident and kept seven years.
Each entry carries a fingerprint of the one before it, so a change to any entry shows. The chained records hold no prompt, answer or person’s words.
EvidenceTraining records for audits
What Happens When a Check Refuses?
When a check refuses an AI action in Syllabium, the action does not run and never reaches a model. The refusal names the check that refused it and is recorded.
Can Review Be Switched Off for AI?
No. In Syllabium, the second-person rule for safety-critical content can be switched off for people’s own changes, never for AI work, and no setting gives an AI agent a plain “allow” on safety-critical content.
Switching the rule off for people’s own changes itself needs a second administrator.
How Do Administrators See AI Proposals?
Syllabium shows administrators a report of how people decide AI proposals, beside an AI activity tab and the AI control pages.
Are Review Decisions Kept for Auditors?
Yes. Every publishing decision in Syllabium is kept in a tamper-evident chain, and closed decisions download as CSV, Markdown or PDF.
AI agents in trainingCoursewareAbout