When an AI answer sounds impressive, can you trust it?
You may already use AI to draft emails, summarise documents, answer customer questions, or prepare internal reports. The output can look polished within seconds. That speed is useful, but it creates a practical concern: how do you know the answer is genuinely reliable, original, and based only on information your business allowed the system to use?
A recent dispute surrounding OpenAI’s claimed solution to the Navier-Stokes problem shows why this question matters. The issue is not only whether an advanced AI model can produce a mathematical proof. It is also about data access, training practices, attribution, independent verification, and the confidence people should place in an impressive result.
TL;DR: AI output should be treated as a draft or recommendation until someone checks it. For your business, clear data rules, human review, source records, and approval steps are essential whenever AI handles sensitive or important work.
What This Means
The Navier-Stokes problem concerns how liquids and gases flow. It has remained unresolved for around 90 years and is one of seven Millennium Prize Problems, each carrying a US$1 million prize for a correct solution. OpenAI said it had found a solution using an internal AI model and 10,000 concurrent agents, according to The Verge.
However, the announcement was followed by questions from mathematicians Tristan Buckmaster and Levent Alpöge. They had published findings on a related problem shortly beforehand and raised concerns about whether their work, entered into AI coding tools, might have influenced the later result. OpenAI said that no specific user data was accessed, while also stating that it could not rule out the possibility that de-identified data from product usage had helped improve its models.
This does not automatically prove wrongdoing. It does show that an AI result can be technically impressive while still requiring careful questions about its origin and verification. A model may produce a plausible answer, but plausibility is not the same as proof. The same principle applies to a business report, sales forecast, compliance summary, or customer response generated by AI.
“A confident AI answer is not the same as a verified business decision.”
For a small business, the practical lesson is simple: you need to know what data went into the system, what the system produced, who checked it, and whether you can explain the decision later.
How This Applies to Malaysian SMEs
Imagine you run a distribution company in Selangor. Your team asks an AI tool to review several months of sales records and suggest which products to reorder. The recommendation may appear sensible, but it could be based on incomplete entries, duplicated invoices, seasonal demand, or a misunderstanding of product codes. If the tool has not been checked against your actual stock records, the result is only a suggestion.
Now consider a professional services firm in Kuala Lumpur preparing a proposal for a client. An AI assistant may draft a strong-looking document, but it could include claims copied from publicly available material, outdated regulatory references, or confidential details from a previous prompt. You need a process that records which information was supplied and requires a person to review every important statement before the proposal leaves your company.
A restaurant, retailer, or online seller faces a similar issue when using AI for customer service. If a chatbot incorrectly promises a refund, delivery date, warranty outcome, or product feature, your staff may need to resolve the complaint. The risk increases when customer conversations contain names, phone numbers, addresses, order histories, or payment-related information. You should decide in advance which information may be entered into an AI tool and which must stay inside your approved business systems.
For Malaysian SMEs working with suppliers, contractors, or external agencies, ownership and attribution also matter. If AI helps prepare marketing copy, product descriptions, designs, or software instructions, keep a basic record of the source materials and human approvals. This helps you answer questions about originality and makes it easier to correct an error when a customer or partner challenges the content.
You should also consider Malaysia’s privacy obligations. The Personal Data Protection Act 2010 regulates the processing of personal data in commercial transactions. This does not mean you cannot use AI. It means you should understand what data is being processed, why it is needed, who can access it, and how your provider handles it.
A simple control system for AI-assisted work
| Control | What you should record | Example |
|---|---|---|
| Input | Information supplied to the AI tool | Approved product list without customer phone numbers |
| Output | AI-generated answer or recommendation | Draft reorder list with assumptions noted |
| Review | Name or role of the person who checked it | Operations supervisor verifies stock figures |
| Approval | Who authorised the final action | Business owner approves supplier order |
| Retention | Where the record is stored and for how long | Versioned folder in the company document system |
This table is not a complicated technical framework. It is a practical audit trail. It helps you avoid the “the AI said so” problem when something goes wrong.
Practical Takeaways
- Classify your work. Mark tasks as low, medium, or high importance. A social media caption may need light review; a contract summary or tax-related document needs stronger checking.
- Keep sensitive data out by default. Remove identification numbers, personal contact details, bank information, passwords, and confidential customer records unless you have approved controls.
- Ask for sources. When AI provides facts, request references and check the original material yourself. For legal or regulatory information, confirm with the relevant Malaysian authority or qualified adviser.
- Separate drafting from approval. Let AI prepare a first version, but assign a person to verify facts, figures, tone, and confidentiality before release.
- Record the prompt and final decision. This makes it easier to investigate mistakes and improve your internal process.
- Use access controls. Staff should only use approved tools and should know which business information may be entered.
- Test before expanding. Begin with one repeatable workflow, measure the errors, and improve the instructions before applying AI across the company.
- Do not claim certainty too early. If the evidence is incomplete, label the output as a draft, estimate, or recommendation.
The Bigger Picture
The OpenAI dispute highlights a long-term change in how businesses evaluate technology. Previously, you may have asked whether software could perform a task. Now you also need to ask whether its answer can be traced, checked, explained, and defended.
This will become increasingly important as AI systems work across documents, customer conversations, databases, and operational tools. The strongest businesses will not simply use the most powerful model. They will build dependable routines around it: limited access, clean data, documented instructions, human review, and clear accountability.
For a Malaysian SME, this is good news because responsible AI use does not require a large research department. You can start with a one-page policy, an approved-tool list, a review checklist, and a shared record of important AI-assisted decisions. These small controls help your team gain speed without giving up judgment.
The key question is not whether AI can produce a remarkable answer. It is whether you can confidently explain how that answer was produced and why your business acted on it. When you can answer both questions, AI becomes a useful assistant rather than an unexplained source of risk.
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 →