Meta’s New AI Deal Could Change How You Share Business Data
Meta is testing a new approach to AI development: instead of simply asking companies to allow their prompts and model outputs to improve future systems, it is offering significantly lower usage rates to businesses that agree to contribute this data. According to TechCrunch, the arrangement applies to Meta’s Muse Spark model, designed for coding and other “agentic” tasks that can perform multi-step work.
For a Malaysian SME, this is more than a pricing story. It is a warning that your everyday interactions with AI may become valuable training material. Customer enquiries, sales scripts, stock reports, internal procedures, source code and financial documents can all reveal how your business operates. Before you accept any data-sharing agreement, you need to know exactly what is being shared, who can access it and whether your team has permission to submit it.
What Happened
Meta is offering a contributor pricing tier for users who permit the company to use their prompts and model outputs to help develop future models. TechCrunch reports that one million input tokens cost US$1.25 under a standard agreement, compared with US$0.10 under the contributor arrangement. Output tokens are reported at US$4.25 per million under the standard pricing and US$0.20 under the contributor pricing. These figures are from the source article and may change, so you should check the provider’s latest terms before making a decision.
The offer comes as AI companies seek more real-world data for improving agentic systems. Unlike a basic chatbot that answers one question, an agent may plan tasks, use software, inspect files and take actions across several steps. TechCrunch cites developer Mario Zechner, who said stored coding-agent sessions helped drive improvements in coding-agent capabilities. The same article also reports that Meta paused an internal employee computer-usage tracking initiative after criticism, showing how sensitive this type of data can be.
The challenge is not limited to software development. AI providers need examples of how people complete professional work, but much of that work takes place inside private systems. A small business may handle quotation requests through WhatsApp, coordinate deliveries in spreadsheets, approve purchases by email and record customer details in a cloud application. Those workflows can be useful for improving AI, but they can also contain confidential information.
Why This Matters for Malaysian SMEs
Many Malaysian SMEs are already experimenting with AI for practical tasks: replying to customer messages in Bahasa Malaysia and English, preparing quotations, summarising meetings, writing social media posts, checking inventory and creating simple automation flows. A contributor plan may appear attractive because it lowers the barrier to testing an AI agent across these activities. However, the data used in a test may include more than the question you typed.
For example, a prompt asking an AI assistant to draft a quotation could include a customer’s name, phone number, delivery address, product requirements and negotiated terms. A request to summarise sales performance could reveal monthly revenue, margins, outstanding payments or supplier details. A coding assistant may receive application keys, database structures or customer records if your staff copy and paste carelessly. Even if the model provider has good security, sharing remains a governance decision for your business.
Malaysia’s Personal Data Protection Act 2010, administered by the Personal Data Protection Commissioner, governs the processing of personal data in commercial transactions. You can review the official guidance and legislation through the Personal Data Protection Commissioner’s website. The Malaysian Government’s MyGOV portal also provides public information on government services and digital matters. Your business should obtain professional advice for specific compliance questions, especially where customer, employee or cross-border data is involved.
The practical issue is permission. Your customer may have agreed for you to use their details to deliver an order, but that does not automatically mean you can send those details to an AI provider for model improvement. Similarly, an employee may use an AI tool for work, but your company should define what information may be entered and which tools are approved.
| Business area | Useful AI experiment | Data to protect |
|---|---|---|
| Customer service | Draft replies and classify enquiries | Names, phone numbers, addresses and complaint details |
| Sales | Prepare quotations and follow-up messages | Customer pricing, margins and contract terms |
| Operations | Summarise stock or delivery information | Supplier terms, inventory levels and route details |
| Software | Review code and suggest fixes | Passwords, API keys, databases and proprietary logic |
What You Should Check Before Joining a Contributor Plan
Do not treat a discounted AI service as a normal software subscription. Read the data-use section, retention period, deletion process, training rights and account controls. Confirm whether prompts and outputs are used to train models, whether your data is separated from other customers’ data and whether you can revoke permission later. Also check whether the provider can use subcontractors or transfer data to other countries.
Ask whether the service offers administrator controls. You should be able to restrict who may use the contributor tier, create separate workspaces for experiments and remove access when an employee leaves. Keep business-critical work outside the experiment until you understand the provider’s controls.
Cheaper AI access is not free if the information you share gives away your customer relationships, operating methods or competitive advantage.
A sensible starting point is to create a “safe data” category. Use fictional customer names, synthetic orders, public product information and anonymised examples. Remove phone numbers, identity-card details, bank information, passwords, signatures and exact addresses. For internal documents, replace actual figures with dummy values while testing the workflow. Once the process works, ask whether you need real data at all.
The Bigger Picture
Meta’s approach shows that data rights are becoming part of the AI product itself. Providers are competing not only through model quality, but also through usage rates, context limits, integration tools and access to real-world feedback. TechCrunch also reports that other frontier AI companies have been lowering selected token costs, indicating an increasingly competitive market. For you, that means the cheapest headline rate should not be the only factor in a purchasing decision.
Agentic AI will likely become more useful when it understands complete workflows rather than isolated questions. An assistant that can read an order, check stock, prepare a delivery note and update a system could save staff time. But each additional action creates another point where permissions, errors and confidential information must be controlled.
Set a simple internal policy before your team starts. Identify approved tools, prohibited data, permitted use cases and who reviews outputs. Require human approval before an AI agent sends a customer message, changes a record, issues a quotation or makes a purchase request. Keep an audit trail of important actions, and review the policy whenever a provider changes its training or retention terms.
For a Malaysian SME, the best response is not to reject AI or share everything. Start with low-risk processes, measure the results and keep ownership of your information clear. If a provider offers lower rates in exchange for your prompts and outputs, treat that offer as a data partnership. Read the terms, protect personal information and decide whether the operational benefit is worth the information you are allowing outside your business.
Ready to Streamline Your Operations?
Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →