What The EU AI Act's Transparency Rule Does To A Singapore Press Release From August
By Gabriel Tan | June 2026
You are an investor relations consultant in Singapore. A listed client sends its half-year results through on a Friday night, and the announcement has to reach the Singapore Exchange by Monday. You open your AI tool, feed it the numbers and last quarter's release, and you have a clean first draft in fifteen minutes. The client has shareholders in Frankfurt and Paris. You do not think about that while you draft. From August, you will have to.
The European Union's Artificial Intelligence Act, the first broad AI law in force anywhere, sets out transparency obligations that apply from 2 August 2026. The reach goes well past Europe. The Act applies to anyone whose AI output touches people in the EU, regardless of where the firm is based. One client with European shareholders is enough to pull your work into range.
Communications firms read “EU law” and file it as a Brussels problem. It becomes a Singapore problem the moment a client's share register carries European names.
The rule lands on synthetic media, not on plain text
Strip the Act down to the part that reaches a communications desk and three duties remain.
First, content that AI generates or alters and then shows to the public can need to be disclosed as such. Article 50, the transparency provision, is written for synthetic and manipulated media: an image no camera took, a voice no microphone recorded, a video of an executive that was assembled rather than filmed.
Second, the disclosure has to be plain to the person reading or watching. A note buried in file metadata does not satisfy it. The reader has to be able to tell that a machine made or changed what is in front of them.
Third, the duty can attach to you. The Act assigns roles, and the party putting content into use under its own name carries obligations, not only the company that built the model. If your firm publishes the output, your firm is in the frame.
The plain release is the easy case
A text announcement carries low exposure. The harder cases are the ones moving into investor communications right now.
A founder video where the audio was cleaned and partly regenerated. A results explainer read by an AI-built avatar. A social card with an image no photographer shot. Each is the kind of synthetic or altered media the rule was written for, and each is becoming routine in the push to produce more for less.
The test is not whether AI helped. The test is whether the audience is looking at something a machine produced or changed, and whether a reasonable person would want to know that. The weight behind the test is real: fines under the Act run up to 35 million euros or 7 percent of global annual turnover. Your client wears that risk, and your firm sits next to it.
What to do this week
You do not need a compliance project. You need an hour.
List every client touchpoint where AI generates or alters something the public sees: images, audio, video, synthetic graphics. Thirty minutes, one sheet.
Mark which clients have any European nexus, whether shareholders, operations, listings, or customers in the EU. Fifteen minutes against your client list.
Write one default disclosure line your team uses whenever AI-generated media goes out, and a rule for where it sits so the reader actually sees it. Twenty minutes.
That is enough to know your exposure and to head off the worst version of this, which is learning in August that a client's investor video needed a label it never had.
The date does not move and the reach does not shrink. A firm that writes its disclosure standard now treats August as a non-event. A firm that waits will be rewriting its PR workflow in the same week its clients start asking what changed. The quiet advantage goes to whoever did the boring hour first.
Gabriel Tan is the founder of Mekong Bridge Advisory. He builds structured execution systems for PR and communications firms.
