When Information Becomes Training Material

You may not be buying rare books, building language models or running a warehouse in Las Vegas. Yet the story about Amazon acquiring rare books, removing their spines and scanning the pages for artificial intelligence training raises a practical question for your business: what happens to the information your company creates and stores?

For a Malaysian SME, valuable knowledge is often hidden in quotation files, customer messages, operating procedures, product catalogues, service reports, invoices and staff know-how. As AI tools become better at reading and summarising documents, that information can help you work faster. It can also create risks if you do not know where your data goes, who can access it or whether you have permission to use it.

The source report says Amazon bought rare books through commercial channels and scanned them to improve its products and services. It also explains why older, hard-to-find text is attractive for AI training: it is less likely to contain AI-generated writing and may not already be available online. Read the original report from TechCrunch.

TL;DR

AI systems need large amounts of reliable text, so businesses are placing greater value on documents that are accurate, unique and difficult to find online.

Your SME should treat internal documents as business assets: classify them, control access, confirm consent and avoid uploading confidential material into public AI tools without a clear policy.

What This Means

Large language models learn patterns from huge collections of text. They may use books, websites, documentation, code, customer support examples and other material during training or improvement processes. The report describes a concern that AI companies are reaching beyond easily available internet content and looking for rare or out-of-print material.

This matters because unique information is more valuable than generic information. A general description of how to write an email is common. Your company’s process for handling a delayed shipment from Port Klang, checking a halal-related product requirement or managing a repeat customer’s preferences is specific. That information may give an AI system useful context, but it may also expose your competitive advantage.

The article also mentions “model collapse”, a situation in which training on too much AI-generated material can reduce the quality of later outputs. The underlying lesson is simple: source quality matters. If you use AI to create customer replies, product descriptions or internal instructions, you still need human review and dependable original references.

Key insight: If a document helps your team make decisions, it should be handled as a business asset before it is treated as convenient input for an AI tool.

How This Applies to Malaysian SMEs

Imagine you operate a small trading company in Selangor. Your sales team keeps product specifications, supplier terms and customer objections in shared folders and messaging apps. One employee pastes a customer’s request into a free AI chatbot to draft a response. The draft may be useful, but the original message could include personal information, internal pricing, supplier details or contract conditions. You may not know how that tool stores or processes the input.

If you run a service business, the same issue appears in job sheets and WhatsApp conversations. A maintenance company may have years of notes about recurring faults, machine settings and customer premises. A renovation firm may store drawings, access instructions and photographs. A clinic-related supplier may handle sensitive records even if it is not a healthcare provider. Before using AI to summarise these materials, you need a clear rule about which information may be uploaded, which must be removed and which should stay inside your own system.

For retailers and food businesses, original content also has operational value. Your recipes, preparation steps, supplier evaluation notes, customer feedback and campaign performance can help an assistant suggest improvements. However, you should separate general marketing copy from restricted information. A staff member can ask AI to rewrite a public product description, but should not paste an unreleased product formula or a complete customer database into the same tool.

Malaysian businesses also need to consider personal data responsibilities. Malaysia’s Personal Data Protection Act 2010 regulates the processing of personal data in commercial transactions, and the official Personal Data Protection Department provides related guidance and resources. Refer to the Personal Data Protection Department. If you use customer names, phone numbers, identification details or purchase histories in an AI workflow, document why the data is needed and restrict access to the people and systems that require it.

Language is another practical consideration. Many Malaysian SMEs work across English, Bahasa Malaysia, Mandarin or Tamil. An AI tool may produce a fluent response while missing local context, formal requirements or the intended tone. Keep approved terminology, translations and standard replies in a controlled knowledge base. That gives your team a dependable reference instead of asking different tools to guess each time.

A Simple Data Classification Table

You do not need a complicated technical programme to begin. Start by sorting common information into clear categories:

Information type	Example	Suggested AI handling
Public	Published opening hours or product features	Suitable for approved drafting tools, with human review
Internal	Staff procedures and sales scripts	Use only in approved business accounts or private systems
Confidential	Supplier terms, quotations and product plans	Do not upload without written approval and access controls
Personal or sensitive	Customer contact details and identification information	Minimise, anonymise or exclude unless a documented purpose exists

The table is a starting framework rather than a legal classification. Your accountant, lawyer or data protection adviser can help you assess specific obligations for your industry and contracts.

Practical Takeaways

• Create a one-page AI usage policy. State which tools staff may use, what information is prohibited and who approves new tools.
• Separate public and private documents. Do not place customer records, supplier agreements and marketing content in one unrestricted folder.
• Remove unnecessary personal data. Before requesting a summary, replace names, phone numbers and account references with neutral labels where possible.
• Check tool settings. Review whether business inputs may be used for service improvement, whether administrators can control access and how information can be deleted.
• Keep a human reviewer. AI-generated text can contain wrong details, invented claims or unsuitable translations.
• Preserve original records. Keep the approved source document so your team can verify what the AI used.
• Record consent and purpose. If customer or employee information is involved, note why it is being processed and who is responsible.
• Train staff with real examples. Show them the difference between asking AI to rewrite a public announcement and uploading a complete customer conversation.

A 30-Day Starting Plan

During the first week, list the tools your team already uses, including chatbots, transcription services, form builders and document assistants. Do not assume that only tools labelled “AI” handle business information.

In the second week, identify your ten most important document groups. These might include customer records, supplier agreements, operating procedures, financial documents and product information. Assign an owner to each group and remove unnecessary access.

During the third week, test one low-risk workflow. For example, use an approved tool to summarise public meeting notes or draft a response based on anonymised feedback. Compare the result with your existing process and record what still needs human checking.

In the fourth week, update your policy, train staff and set a review date. Revisit it whenever you add a new tool, start working with a new partner or change the type of information your business collects.

The Bigger Picture

The rare-book story shows that AI development is not only about software and computing power. It is also about access to high-quality information. As more companies use AI for search, support, sales and operations, distinctive business knowledge will become increasingly important.

That does not mean you should keep every document locked away. It means you should make deliberate choices. Some information can safely improve an internal assistant. Some should be anonymised first. Some should never leave your systems. The strongest SMEs will know the difference.

For you as a business owner, the sensible goal is not to avoid AI. It is to build a controlled path for using it. Start with clean documents, clear ownership, limited access and simple staff rules. When your team understands what may be shared and what must be protected, you can gain practical benefits without casually giving away the knowledge that makes your company different.

Ready to Streamline Your Operations?

Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →