Your AI Tools Might Be Built on Unstable Ground

You trust your AI tools. They write your emails, generate your social media captions, and power your customer chatbots. For a busy Malaysian SME owner, these tools feel like a lifesaver. But have you ever stopped to ask where the data your favourite tool learned from actually came from?

A recent important court ruling in the United States has exposed a significant fault line in the AI industry. It wasn’t about an AI going rogue. It was about a very old-fashioned concept: theft. An AI company was caught training its models on data obtained from pirate websites. The legal consequences were severe, and the ruling sends a clear warning to every business that relies on this technology.

This isn’t just a problem for big tech companies in Silicon Valley. The data used to train the tools you rely on every day might be on shaky legal ground. Understanding this ruling means understanding the true risk of the automation tools you are adopting for your business.

TL;DR

• A US court ruled training AI on copyrighted data can be “fair use,” but how the data was collected (via piracy) was outright illegal. This split ruling created a massive legal liability.
• Malaysian copyright law is stricter than US fair use. If you are using or building AI tools, the source of the training data matters more than ever.
• You must audit your AI software vendors now. Using tools built on questionable data puts your business operations and legal standing at risk.

What This Means in Plain Language

The AI company Anthropic was sued by a group of authors and publishers. They argued the company used their copyrighted books to train its AI models without permission. The court’s decision was a split verdict. The judge agreed that feeding a copyrighted book to an AI to learn from it is broadly considered “fair use” — a key legal protection for AI developers.

However, the judge did not accept how Anthropic built its library. The company had downloaded a significant number of books from known pirate sites like Library Genesis. The judge ruled this was simple, direct copyright theft (TechCrunch). To avoid a trial that could have ended even more severely, the company settled the case. The core lesson is stark: the method of data acquisition is now a critical legal battleground. It doesn’t just matter what the AI does; it matters what it ate to get there.

For you, this means any AI tool that was trained on a questionable internet scrape—especially content aggregated without permission—carries a hidden liability. The vendor faces the risk, but the instability that creates becomes your operational problem.

How This Applies to Malaysian SMEs

It is easy to think a US court ruling only affects American companies. That would be a mistake. The AI tools you use daily—from ChatGPT to Midjourney to various productivity automations—are primarily built by US companies. Their legal problems inevitably become your operational problems.

Your Legal Exposure is Different. Malaysia operates under the fair dealing doctrine, which is significantly narrower than the US “fair use” standard (MCMC). If a tool is built on data that is even slightly unstable in the US, it is almost certainly unstable here. If your SME uses AI to generate marketing materials or documents, and those AIs rely on pirated data, you are building your brand on an unlicensed foundation. If the AI provider is forced to change its model, delete data, or shut down, your operations stop.

The “Clean Training Data” Advantage. This ruling creates a clear separation in the market. On one side, you have AI developers who carefully license their training data (like Adobe with Firefly, or Microsoft with Copilot). On the other side, you have developers who scraped the open internet without clear permissions. For a Malaysian SME, choosing the “clean” option is not just about ethics; it is about operational safety. You want a tool that will be around in a year, with a clear legal path forward.

Your Internal Automation Projects Are at Risk. Many SMEs are starting to build their own simple automations: chatbots for customer service, or internal knowledge bases using simple AI. This ruling is a direct warning to you. If you scrape data from the web, from competitor websites, or from social media to feed your internal AI, you might be building the exact same legal trap for your business. The Anthropic case makes it clear that the data source is the critical point of failure, no matter how big or small your project.

Practical Takeaways for Your Business

• Audit your entire AI software stack. Make a list of every AI-powered tool your team uses. Google, Meta, OpenAI, and independent apps all have different data sourcing policies.
• Check for Indemnification. Read the terms of service. Does your AI vendor promise to cover your legal costs if their tool is accused of copyright infringement? Microsoft and Adobe do for their commercial products. Many popular tools do not.
• Stop Pasted Data. Do not copy-paste your proprietary documents, client lists, or copyrighted industry reports into public AI chatbots. You are effectively exposing your own data to the same scrutiny the author’s books were exposed to.
• Build Your Data Moats. If you are automating processes, use only your own data. An internal database your company generated is much safer than a summary derived from a public web scraper.
• Follow the Local Cases. Malaysian law firms and IP bodies are watching these rulings closely (Skrine Legal Updates). Ensure your compliance approach aligns with local expectations, not just US ones.

The Bigger Picture

This ruling is a defining moment. It signals the end of the “wild west” period of AI development where data was simply vacuumed from the internet. We are moving from an era of “model size” to an era of “data provenance.” The value of an AI model will no longer just be what it can do, but how it learned.

For an SME in Malaysia, this is a strategic inflection point. The tools you choose today should be evaluated for their legal and data resilience, not just their feature list. The safest path is to align with AI providers who demonstrate clear, transparent data sourcing. As the legal net tightens globally, and as local laws adapt, the businesses that invested in “clean” AI will be the ones that don’t have to start over completely.

“If the AI you depend on for your business operations learned its skills from stolen material, your business is sharing that liability. You cannot have stability in the output if there is no integrity in the input.”

AI Type	Typical Data Source Risk	Best Action for Your SME
General Chatbot (ChatGPT, Gemini)	Medium (Web scrape, opt-out)	Use for drafts, do not paste sensitive data.
Image Generator (Midjourney, DALL-E)	Higher (Public art scrape)	Use for ideas. Avoid for final commercial assets.
Enterprise AI (Copilot, Firefly)	Low (Licensed / Indemnified)	Best for internal documents and core business outputs.
Open Source AI (Llama, Stable Diffusion)	Highest (Unknown sourcing)	Require deep legal vetting before business use.

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 →