AI-Generated Text Is Becoming Easier to Identify
If you use AI to draft product descriptions, customer replies, proposals, social media posts or internal documents, a major change is taking shape: AI-generated writing may soon carry an invisible signal that helps identify how it was produced. Anthropic has explained that its Claude chatbot will use a text watermarking approach based on Google DeepMind’s SynthID-Text technology. The signal is designed to remain invisible to readers while remaining detectable through the right verification system. Source: The Verge
For a Malaysian SME, this is not just a story about large technology companies or European regulation. It raises practical questions about trust, content approval and record-keeping. If you use AI in your business, you should know when content is machine-assisted, who reviewed it and whether your customers, business partners or regulators may expect greater transparency in the future.
What Happened
Anthropic says Claude’s invisible text-marking system will be “a version” of SynthID-Text, an open-source technology developed by Google DeepMind. The method works by influencing small word-selection decisions made during text generation. When several words would fit naturally in a sentence, the system uses a special key and surrounding words to guide the selection rather than relying on an ordinary random process. Source: The Verge
For example, after a phrase such as “The weather today was cold and…”, words such as “overcast” or “grey” may fit the meaning, while “sugary” would not. The watermarking system works through many of these low-impact choices across a longer piece of writing. The resulting pattern is not meant to be visible to the reader, but it can be detected by someone with the relevant key. Anthropic says the system is intended not to create a practical change in the quality or content of Claude’s responses. Source: The Verge
The move is connected to the European Union’s AI Act, which includes transparency requirements for synthetic audio, images, video and text. Anthropic is also introducing C2PA support for images processed by Claude, while Google’s Gemini has supported SynthID Text since 2024. OpenAI has not publicly detailed a text-watermarking plan in the cited compliance roadmap, although the article notes that it will also be subject to the EU requirements. Source: The Verge
AI assistance may become invisible to the reader, but it should not become invisible to your business process.
Why This Matters for Malaysian SMEs
Most Malaysian SMEs do not need to redesign their operations immediately because a technology company is adding watermarks. However, you do need to consider how AI-generated content affects your reputation. A café may use AI to prepare bilingual menu descriptions. A local wholesaler may use it to answer enquiries from retailers. A software reseller may use it to prepare product comparisons. A recruitment agency may use it to draft job advertisements. In each case, the final responsibility still sits with your business.
An invisible watermark will not prove that a statement is accurate, original or approved. It only helps indicate that a text-generation system may have been involved. You should therefore treat AI detection as one part of governance rather than as a replacement for human review. A staff member should still check claims, names, dates, product specifications, customer details and language before anything is published or sent.
This is especially relevant when you communicate in English, Bahasa Malaysia, Mandarin or Tamil. AI tools may produce fluent text that sounds appropriate but misses local context, customer expectations or industry terminology. A watermark may help identify AI involvement, but it cannot tell you whether a Malay-language promotion is culturally suitable or whether a technical explanation is safe to rely on.
If you serve overseas customers, sell through international platforms or work with companies that have European operations, transparency expectations may reach your workflow indirectly. The EU AI Act is the reason cited by Anthropic for introducing these markings, but Malaysian businesses may still encounter clients or platforms that request disclosure of AI-assisted content. Source: The Verge
Practical Use Cases for Your Business
| Business activity | What you should do | Why it matters |
|---|---|---|
| Customer service replies | Let AI prepare a draft, then require staff approval before sending. | AI detection does not confirm that a reply is correct or suitable. |
| Marketing and social posts | Keep a simple record of the tool used and the person who approved the post. | It creates an internal trail if a customer questions the content. |
| Product descriptions | Compare every generated claim against your catalogue or official specification. | Fluent writing can still contain inaccurate features or promises. |
| Business proposals | Mark AI-generated sections internally and review confidential information. | It helps prevent sensitive details from being reused without approval. |
| Images and promotional materials | Check whether the platform supports provenance or content credentials. | C2PA support is being used by Anthropic for Claude-processed images. Source |
What You Can Do Now
First, create a short AI-use policy written in plain language. It does not need to be a lengthy legal document. State which tasks employees may use AI for, which information must not be entered into public tools, who checks the output and when customers should be informed that content was AI-assisted.
Second, introduce an approval label in your workflow. A document can be marked “AI draft”, “human reviewed” or “approved for external use”. This is useful even if your chosen AI tool does not currently watermark text. It separates drafting from publishing and makes responsibility clear.
Third, protect confidential information. Do not paste customer identification details, private contracts, unreleased product plans, passwords or sensitive financial records into an AI system unless you have checked the provider’s terms and your own data-protection requirements. Malaysia’s Personal Data Protection Act 2010 regulates the processing of personal data in commercial transactions, so your AI workflow should be reviewed alongside your existing privacy practices. Source: Malaysia Personal Data Protection Commissioner
Fourth, keep a human-readable record of important AI-assisted work. Record the date, tool, purpose, reviewer and final version for documents that affect customers, employees or business partners. This is a practical control you can implement regardless of whether an invisible watermark can be detected later.
The Bigger Picture
Text watermarking is part of a wider shift towards content provenance. As AI-generated writing, images and other media become common, businesses will need better ways to explain where content came from and how it was checked. Anthropic’s approach uses patterns in word selection, while its image plans reference C2PA support. Google’s SynthID Text is another example of a provider developing machine-detectable signals. Source: The Verge
For you, the important question is not whether every customer will inspect a watermark. The practical question is whether your company can stand behind the content it publishes. A supplier may ask whether a proposal was prepared with AI. A customer may question an unusual product claim. An employee may rely on an AI-generated instruction. A clear review process gives you an answer that technology alone cannot provide.
Do not treat watermarks as a guarantee of authenticity. They may be difficult to detect after heavy editing, translation or rewriting, and an unmarked document is not automatically written by a person. Instead, combine provenance signals with staff review, source checking, privacy controls and clear approval steps.
AI tools are becoming part of everyday SME work, from responding to enquiries to preparing campaigns. The businesses that benefit most will not simply generate more content. They will build dependable habits around it. Start by documenting your current AI use, deciding which work requires human approval and telling your team that speed never removes accountability.
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 →