The Monthly Ramble · AI

The Honest Label

Our first experiment with AI was this website and a little game, and we're crediting the tool that helped. Here's why being open about AI matters, and why clients are really paying for a person's judgment.

Cliff, Founder·November 2026·9 min read·AI Brand Law

An orange paper tag reading HONEST hanging from a thin black string on cream

A small confession to start

I'll go first. The RegenLFG website you're reading right now was built with help from Grok Bot, an AI assistant. So was Headline Smack, the little game tucked into our Opinions section. People on our team steered both, checked the work and made the calls on what stayed.

The site and the game were our team's first real experiment with AI. Every piece of client work you see on it, the summits, the films, the illustrations, the emails, the articles, was made by people, without AI. Writers, designers, editors and producers, people who care about getting the small details right.

I'm telling you this not because anyone asked, but because it's credit where it's due. If a production partner helped us shoot a film, we'd name them in the credits. Grok Bot helped build the site and the game, so it gets named too. And if we ever bring AI into client work down the road, we'll say so just as plainly.

"We use AI" is not a pitch

You've probably seen agency decks where the big reveal is "we use AI." I understand why. It sounds modern. But to a client it often reads like slop: a promise of more stuff, faster, with nobody quite responsible for it.

Clients in fintech, payments and Web3 are not short on content. They're short on trust. They don't hire an agency for volume. They hire for taste, for knowing which headline will get a compliance team to sign off and still make a reader stop scrolling, for knowing when a film needs to be ninety seconds and not three minutes. Tools can speed parts of that up. They can't do the choosing.

That's why our proposition has never been "with AI." It's people who love to experiment with tech. We try new tools, test them properly, keep what earns its place, drop what doesn't, and we're open about which is which. Building this site and Headline Smack with Grok Bot is exactly that kind of experiment. The curiosity is the point. The tool is just the tool.

It's a bit like trail running. Good shoes help, but they don't finish the race for you. What gets you down a steep, rocky trail is reading the ground, and you only learn that with practice.

When to tell people

My rough rule is simple: disclose whenever the answer would change how someone feels about the work.

For clients, that means telling them up front if AI is part of the process, what it's used for, and what it isn't. Put it in the proposal, not in a footnote after delivery. If we ever generate an image, synthesize a voice, or use a tool on their data, they should know before it happens and get to say no.

For audiences, it means labelling anything that could mislead: a synthetic face, a cloned voice, a scene that never happened. A spellchecker doesn't need a label. A realistic video of a CEO saying something they never said absolutely does.

And when nothing was generated? It's worth saying that too. "Made by people" is becoming a real signal, but only if it's true.

The legal side, as a list of things to check

I'm not a lawyer and this isn't legal advice. Rules are moving fast, so treat this as a snapshot as of early November 2026 and a list of questions to put to your own counsel.

Singapore

Who owns it? Copyright still rests on human authorship. The Court of Appeal tied copyright to identifiable human authors in Global Yellow Pages v Promedia (2017). Section 244 of the Copyright Act 2021 allows copying for computational data analysis, which covers training, if access is lawful, but that exception doesn't cover what you later publish. MinLaw and IPOS published a public consultation on AI and intellectual property in August 2026, so check whether anything changes. If a client expects to own and enforce a deliverable, check how much of it a person actually made. Whose face and voice is it? The Online Safety Commission began operations on 29 June 2026 under the Online Safety (Relief and Accountability) Act 2025, which Parliament passed on 5 November 2025. Its first phase covers five harms, such as harassment and intimate image abuse. "Inauthentic material abuse," the deepfake category, will come later; MDDI said on 10 September 2026 that dates would follow. In the same answer, MDDI noted that under the Singapore Code of Advertising Practice, advertisers should disclose AI use where needed to avoid misleading people and shouldn't portray a person without consent. It also noted new Online Criminal Harms Act codes from 17 August 2026 that require designated platforms to verify advertisers. Since 22 January 2025, realistic deepfakes of election candidates in online election ads have been banned. Get written consent for any synthetic use of a real person. Whose data is it, and who's in charge? Feeding client or customer information into an AI tool can raise PDPA questions. The PDPC's Advisory Guidelines on personal data in AI systems (1 March 2024) are still the place to start. If you're trying AI agents, IMDA's Model AI Governance Framework for Agentic AI launched on 22 January 2026, and version 1.5 came out on 20 May 2026. It's voluntary, but its core idea fits marketing well: keep humans meaningfully accountable and tell people when agents are involved.

United States

The Copyright Office said in January 2025 (Part 2) that purely AI generated material isn't protected. On 9 May 2025 it released a prepublication Part 3 on training, which treats fair use as a fact by fact question. Two 2025 rulings in California, Bartz v Anthropic and Kadrey v Meta, found training on books could be fair use. Bartz drew a line at pirated copies, and Kadrey left the door open to "market dilution" arguments. On disclosure, California's AI Transparency Act became operative on 2 August 2026 and requires covered generative AI providers to embed provenance data and offer free detection tools. Meanwhile, an 11 December 2025 executive order directs federal challenges to state AI laws, but on its own it doesn't override them. If you work with US clients, check which state rules touch your campaigns.

European Union

The AI Act's Article 50 transparency rules have applied since 2 August 2026. Chatbots must say they're AI, deepfakes must be labelled, and generative tools must mark output in a machine readable way. The Digital Omnibus delayed the high risk rules to December 2027 but left Article 50 in place, with a grace period to 2 December 2026 only for marking by systems already on the market. The Commission published Article 50 guidelines on 20 July 2026, and a voluntary Code of Practice on transparency of AI generated content followed on 31 July 2026. If a campaign reaches people in Europe, check whether you're in scope.

Elsewhere in Asia and Australia

China: The measures on labelling AI generated content took effect on 1 September 2025, and they're being enforced. On 28 April 2026, the CAC announced action against platforms, including the CapCut app (剪映), over labelling failures. An amended Cybersecurity Law with AI provisions took effect on 1 January 2026. South Korea: The AI Basic Act and its enforcement decree took effect on 22 January 2026. They include labelling duties for generative AI output, and MSIT announced a grace period of at least a year on fines. Japan and Vietnam: Japan's AI Promotion Act fully took effect on 1 September 2025; it's a promotion framework, and existing copyright and consumer rules still apply. Vietnam's AI Law (No. 134/2025/QH15) took effect on 1 March 2026. Australia: On 26 October 2025, the Attorney General ruled out a text and data mining exception for AI training. Malaysia, Indonesia, Thailand: As of late October 2026, they still have no comprehensive AI law in force, though drafts are in motion. Keep an eye on them.

The common thread: label what could mislead, get consent for faces and voices, keep records of what a person made, and read the tool's terms. None of that is hard, and all of it is easier to do before the work starts.

Where I land

None of this is about pretending AI doesn't exist or swearing it off. We've started experimenting with it, and Grok Bot helping build this site and the game is the first thing I'm happy to credit. What I'm against is confusion: work where nobody can tell what's real, who made it, or who stands behind it.

So here's what we're trying to do. Say what we use, and where. Ask before we touch anyone's face, voice or data. And keep the judgment where it belongs, with people who'll put their name on it.

That feels like the honest label. I'd love to hear how your team handles it.

The Honest Label

A small confession to start

"We use AI" is not a pitch

When to tell people

The legal side, as a list of things to check

Singapore

United States

European Union

Elsewhere in Asia and Australia

Where I land

Sources

Related reading and work

Cliff

More rambles

The sticker leaves the store. That's the point.

You can't buy your way into someone's personal shopper

Reddit's bad week says it plainly: GEO is here to stay

Want to talk about how your team labels its work?