This Sunday, Singapore’s Prime Minister Lawrence Wong will deliver his third National Day Rally in English, Malay and Mandarin.
I do not claim to know how the speech was prepared. But it would hardly be surprising if an AI agent retrieved policy papers or checked claims, while another agent may have helped with translation. After all, PM Wong chairs Singapore’s National AI Council, created to accelerate responsible AI use across the economy.
Would those AI tools make this year’s NDR speech any less his? No. Authorship rests on control of the policy argument and taking responsibility for the result. The PM decides what is delivered and committed to at the NDR.
Now, if that principle holds for the PM, should it hold for other professions too?
Watermarking rules risk regressing AI adoption
Europe’s new AI rules now put that principle under pressure. Article 50 of the EU AI Act has applied since August 2, 2026. It requires providers of certain generative AI systems to mark synthetic output in machine-readable form. A limited grace period until December 2 applies to the marking obligation for systems already placed on the market before August 2.
Certain AI-generated public-interest content also requires disclosure when there is no human review or editorial control, while editing that does not substantially alter the user’s input is exempt from the provider marking requirement. Breaches can attract fines of €15 million or 3 percent of a company’s worldwide turnover.
Global AI model providers can carry those rules far beyond Europe through their commercial and product decisions.
Anthropic, one of the world’s most prominent private AI companies, has introduced an invisible watermark for text generated by newer Claude models, with the marking applied globally. Yet to the ordinary reader, a detected mark indicates only the likelihood that Claude was involved in producing the text. It cannot show whether a writer accepted minor edits or simply handed over the entire argument.
OpenAI also developed and tested a text watermark in 2024 but did not deploy it. Research reported at the time found that nearly 30 percent of ChatGPT users would use the product less if watermarking were introduced.
Singapore may not have imposed any such rule, but Anthropic’s policy still applies here. That raises two possible consequences for individuals and businesses that deserve closer examination.
First, the price of regression.
A Slack survey of 17,372 office workers across 15 countries, including Singapore, found almost half were uncomfortable admitting AI use to their managers. Among those who were uncomfortable doing so, 47 percent felt that using AI was “cheating,” while 46 percent feared being seen as less competent and the same proportion feared appearing lazy.
The same stigma can appear in consumer choices. A 2024 study involving more than 1,000 US adults found that explicitly describing products as using artificial intelligence reduced purchase intentions across a range of products and services.
Yet in reality, professional work already combines human judgment with machine assistance. Radiologists use software to identify abnormalities; genomic researchers use it to find patterns faster; software developers incorporate AI-assisted code and remain responsible for testing it; filmmakers who use AI to plan their scenes do not surrender creative control.
We do not routinely place warnings on the finished work of these professions. Shouldn’t writers and illustrators, whom Anthropic’s policy most directly affects, deserve the same presumption of competence?
Taken further, if workers anticipate that ordinary AI use will invite suspicion about their competence or integrity, some will understandably avoid it altogether. And if that avoidance concentrates in exactly the professions where detection happens to be easiest, hiring managers, clients and institutions could start favoring professions that carry no detectable AI marker over those that do, regardless of the actual quality of the work.
That is a structural distortion in who gets to compete on equal footing, which, in my view, is regressive in our push for broader AI diffusion.
Second, tilting the playing field.
Consider a speechwriter who drafts a speech and asks Claude to tighten several paragraphs. The resulting text may carry Claude’s watermark. Yet if a claim in the speech proves false, Anthropic bears no responsibility for it; the writer does.
This exposes the gap between provenance and authorship. Provenance identifies the system used during production; authorship rests with the person who formed the argument and accepts the consequences of publication.
A watermark can indicate that Claude was involved somewhere in the process, but it cannot measure the extent of that involvement. And because the signal itself carries no standard interpretation, it risks becoming whatever the more powerful party in the room decides it means.
For example, a client who wants to renegotiate a fee or question a writer’s contribution now has a detectable fact to point to, regardless of how much of the argument, research or judgment actually belonged to the writer.
Here is another example of how the playing field could be tilted. Unlike a large communications agency, an independent writer has far less power to challenge a client who treats the mere detection of AI use as evidence that the writer contributed little to the paid work.
I chair PRCA Asia Pacific’s Public Affairs Group, which represents policy and communications professionals across firms large and small. Anthropic’s decision should prompt our industry to scrutinize the standards being set by global AI providers and the uneven burden they may impose.
Clinical researchers and software developers may use AI just as extensively, yet their finished work carries no comparable marker. Subjecting writers, illustrators and content creators to greater scrutiny simply because their output is easier to inspect is plainly unfair.
Who controls the work and stands behind it?
To be sure, regulation has a pivotal role in safeguarding society. Singaporeans lost S$913.1 million to scams in 2025. In one case, a victim transferred at least S$4.9 million after scammers used deepfakes of PM Wong and other government officials during a video call. These are real harms, and regulation should confront the deception behind them.
At the same time, Singapore wants more people to use AI with confidence while remaining responsible for what they produce. Anthropic’s response to Article 50, by watermarking AI-generated text, risks pulling in the opposite direction, however unintentionally, if the use of a legitimate tool itself becomes grounds for suspicion.
Ultimately, the fairer and more rational test is whether a person controls the work and stands behind it.
PM Wong’s NDR speech will remain his because he owns its content and answers to the country for it. Shouldn’t the same test apply equally to a researcher’s report or a developer’s code, as it would to a writer’s article or a filmmaker’s production?
Marcus Loh is the chairman of the Public Affairs Group at PRCA Asia Pacific and a director at Temus, a Singapore AI and digital services firm. He is currently reading War Studies at King’s College London.
Editor’s note: This contributed article has been lightly edited for clarity, length, and style. Where appropriate, TNGlobal may verify, qualify or omit factual claims that cannot be independently corroborated. The views and arguments expressed remain those of the author.
Share your perspective: TNGlobal welcomes contributed insights and expert commentary from across Asia’s technology and innovation ecosystem. Submit a contribution for editorial consideration, or explore more conversations in our TNGlobal INSIDER and TNGlobal Q&A and Interviews archive.
Featured image: Steve A Johnson on Unsplash
My fellow Singaporeans, “Dear You” wasn’t meant for us. What happens when it is?