Most marketing teams now have AI writing the copy, AI choosing the audience, and at least one AI feature nobody remembers turning on.
Ask who knows what those tools do with customer data, and the room goes quiet. Mine did.
That quiet is exactly what regulators have started to notice. AI in marketing has moved from drafting words to making decisions, and the paperwork is catching up.
If you work across the region, this is a calendar I’d pin up regarding AI regulations:
2 December 2026: EU. Watermarking deadline for generative AI tools already on the market. Mostly your vendor’s problem, but worth one email asking them what they’re doing about it.
10 December 2026: Australia. Privacy policies have to explain how personal information is used in automated decisions that could significantly affect someone. It covers decisions made after that date, even if the data was collected years ago.
Late December 2026: India. ASCI’s guidelines on labelling AI-generated content in ads were released on 29 September and take effect three months later.
22 January 2027: South Korea. The grace period on fines under the AI Basic Act ends. The labelling rules themselves have technically applied since January this year.
Singapore: nothing on the calendar. The government said last November it had no plans for specific AI labelling rules on product and service content, and in August said it’s still watching how labelling develops overseas. The one new document is MAS’s AI risk management guidelines, issued on 7 October, which apply to financial institutions from October 2027.
Next, if your ads run in Singapore, Sydney, Seoul, and Mumbai, or just anywhere in APAC where the rules differ starkly from market to market, here are 5 questions you should be asking your legal team.
1. “Which of our tools decide something about a person, and which markets are those people in?”
Start here, because it’s the one most teams can’t answer off the top of their head. A copy tool doesn’t decide anything about anyone. A model that picks who qualifies for an offer, sets a price, or filters who sees a credit card ad does. There’s no single high-risk list to check against in this region, so the most useful test I’ve found is Australia’s: does the tool make the decision, or produce output that’s substantially and directly related to it? If the answer is yes and those people are in Australia, from 10 December your privacy policy has to say so.
2. “For each market our campaigns run in, does an AI-generated person, voice or product shot need a label?”
The answer changes at every border. India’s ASCI now says synthetic influencers and ambassadors must be labelled, while AI-written ad copy doesn’t need one. Korea wants a label on generative output that’s hard to tell from the real thing. China requires both a visible label and one in the file’s metadata. Singapore has no mandatory label, but ASAS applies the same standard to an ad whether or not AI made it. And the line from ASCI I’d underline: a label doesn’t rescue a misleading ad. It’s still banned.
3. “Who specifically checks figures before anything AI-drafted goes live?”
“Someone on the team” isn’t an answer. This one’s the same in every market: ASAS has said plainly that advertisers stay responsible for keeping content truthful, regardless of whether AI was involved. In financial services it’s doubly true. Any number about rates, fees or returns should have a named human checker before it ships.
4. “If a customer asks why they were shown this offer, what do we tell them, and who tells them?”
Most teams have never had to answer this out loud. In Singapore, the document to know is PDPC’s 2024 guidance on AI recommendation systems, which says the PDPA’s consent and notification obligations apply when you use personal data to recommend or decide things for people. In Australia, from 10 December, part of that answer has to sit in your privacy policy. If nobody owns the reply, that’s your gap.
5. “Who signs off on what our agents send, and is their name written down anywhere?”
Because “the AI did it” won’t work with any regulator in the region, and several have said so in writing. ASAS puts responsibility on the advertiser. MAS’s new guidelines say financial institutions remain responsible for risks from third-party AI providers. IMDA’s agentic AI framework is built around humans staying accountable. If the sign-off only lives in someone’s head, it doesn’t really exist.
6. “Which of our tools train on what we feed them, and have we switched it off?”
Worth checking whatever market you’re in. The Singapore twist: PDPC’s position is that training AI on personal data needs consent unless an exception applies. So a vendor quietly training on your customer emails isn’t just a terms-of-service annoyance. It’s a PDPA question.
What to do this week
Don’t send legal six questions. Send them one page.
Three columns: every AI tool your team uses, which markets its output reaches, and whether it decides anything about a person. That page will get you a better conversation because it tells legal where to look instead of asking them to guess.
If you want a head start, paste your tool list into Claude or ChatGPT with this:
“Here’s a list of the AI tools my marketing team uses, what each one does, and the markets our campaigns run in. For each tool, flag: (1) whether it makes or shapes a decision about an individual, (2) whether it produces people, voices or product visuals that appear in ads, and (3) which of those markets might have AI labelling or automated-decision disclosure rules. Don’t tell me what the law says. Tell me which questions to take to our legal team, per market.”
And the obvious caveat, which applies to the original too: I’m not a lawyer. This is a map for a better conversation with yours, not a replacement for it.
The AI;DR
Elsewhere in the AIverse
Disney bids $20M for bankrupt Character.AI. Character.AI, the app where people chat with AI versions of fictional characters, filed for bankruptcy last week, five years after launching. The filing blamed high costs and potential liabilities from lawsuits and regulatory scrutiny, including government investigations over child safety. Disney has offered $20 million for some of its assets and will make job offers to at least 60 staff, most of the company.
Amazon’s Bee launches two AI note-taking wearables. Bee Frontier goes on your wrist or clips to your clothes, at US$74.99. Bee Eclipse is a card-sized device that sticks magnetically to the back of your phone, from US$99.99, and isn’t out yet. Both listen to conversations and turn them into summaries, to-dos and draft follow-up emails, with Gmail, Google Calendar and Outlook built in.
Ramp opens up Glass, its internal AI workspace. Ramp, the US corporate card and spend management company, has opened the AI workspace it built for itself to other businesses through an early access programme. Glass lets teams share reusable workflows, which Ramp calls skills, keep hold of company knowledge, and work on the same task together.