AI LMS

What makes an AI LMS different

"It has AI features" and "it is an AI LMS" are different claims. The second means internal training material and question banks organised into a data layer the AI can reference, with a commercial LLM connected through a controlled path. Seven things to make it do in a demo settle which one you are looking at. Does it cite its grounding, does the source survive, does a person approve the output.

things to make it
do in the demo

AI controls named
in public RFPs

230GB

internal material one public
RFP put in scope for RAG

Request an AI LMS demo The 8 AI controls

Telling them apart

Seven things to make it do in the demo

If what each item describes does not appear on screen, the feature is a demo prop. Tick the box on the right yourself while it runs.

Answers grounded in internal material (RAG)
The answer names the source document and the place in it. Ask something the document does not cover and it says so instead of inventing.
Automated content drafting
One file becomes a draft while the clock runs in front of you. Nothing publishes before a person approves it.
Personalized recommendation
The reason for the recommendation is shown next to it.
Completion prediction & drop-off alerts
It does not stop at a score: it carries through to segmenting the at-risk group and sending the reminder.
Natural-language queries for admins
The answer reconciles against the underlying aggregate metric.
Auto captioning & translation
A person edits and approves, and publication is blocked before approval.
Latency & cost observability
Call volume, latency and cost are visible in the admin console.

Procurement requirements

The 8 AI controls — ready to paste into an RFP

What counts is not how many a vendor says it supports, but how many it can evidence in writing.
Ask every candidate for the right-hand column in this form and the comparison writes itself.

Eight AI control requirements for an AI LMS, the evidence to demand for each, and the published evidence from TouchClass
Control	Evidence to demand	TouchClass
1. Excluded from model training	API configuration record · contract clause · sub-processor scope	Knowledge assets a customer creates or provides are not used as AI model training data.
2. No outbound sensitive data	Filtering policy · masking rules · exception criteria	The AI chatbot answers from internal training material rather than the open internet. File upload and download can be blocked.
3. Limited call scope	Permission matrix · per-feature AI access settings	AI chatbot, ShortClass and QuickMaker rights are granted at sub-administrator level.
4. Input and output validation	Validation rules · blocked examples · hallucination mitigation	Administrators set answer scope and tone by prompt, with per-category scope limits.
5. Source attribution	Attribution screen · path to open the original document	Course Q&A answers display the training material they were drawn from.
6. History and audit logs	Log schedule · retention period · export method	△ Access control and operational records are managed within the scope of ISMS-P and ISO/IEC 27001:2022. An AI-call-level audit log schedule is not covered by any public document.
7. Certified deletion	Data inventory · deletion procedure · disposal certificate	△ Stored in the AWS Seoul region, AES-256 at rest and TLS in transit. The disposal procedure for preprocessing data and embeddings is not covered by any public document.
8. Output control by role	Download permission policy · separated approval rights	Sub-administrator separation and least-privilege access control, with screen-capture blocking and watermarking.

We do not claim to publish what we have not published. Controls 6 and 7 stay △ because no public document evidences them. Hold every other candidate to the same standard.

Putting AI requirements into an RFP — give them their own chapter rather than blending them into the general functional requirements, and demand the eight controls above with their evidence documents. Anything that can only be answered verbally belongs in a contract clause. The drafting procedure, from scope to exit terms, is in the 8 steps of an LMS RFP.

Source: "2026 e-Learning Platform AI Agent Development" request for proposal, Korea Association for Radiation Application, published 2026-05 on Korea's national e-procurement system. It put about 230GB of training material and a 718-item question bank in scope for RAG, and scored 33 requirements 80 technical to 20 price.

Seven comparable 2025–2026 notices from Korean universities and public institutions repeat the same demands.

Verify these on the security, enterprise security, AI assistant and AI admin pages.

FAQ

Frequently asked questions

What is an AI LMS?

Not a learning management system with AI features bolted on. It is internal training material and question banks organised into a data layer that AI can reference, with a commercial LLM connected through a controlled path. The test is not whether AI is present but whether there is grounding, attribution and an approval step.

What should we compare AI LMS vendors on?

Evidence for the eight controls, not the number of AI features. Separating "we support that" from "here is the document" narrows a shortlist quickly. For choosing the LMS itself, see how to choose an enterprise LMS.

What does RAG mean in an LMS?

Internal documents are retrieved first, and the answer is generated from what they contain. Published requests for proposal specify the whole chain, from splitting documents into retrievable units through to source attribution. The first gate is therefore not model choice but the state of your own material.

What does AI learning governance have to cover?

Five components: an AI Gateway governing external calls, a RAG Knowledge Base holding internal material, a Policy Engine for permissions and input/output validation, an Audit Log, and a Review Workflow covering generation, review and approval. That is the published procurement requirement restated as architecture, not one product's feature list.