AI Disclosure Is the New Sustainability — and the Performance Has Already Started

There is a sentence appearing in annual reports this year that almost no one wrote five years ago: we are deploying artificial intelligence across our operations to drive efficiency and growth. It is vague, it is upbeat, and for now it is entirely unverifiable. Anyone who lived through the first wave of sustainability reporting will recognise the genre at once. We have seen this film, and we know how it ends.

It ends with disclosure regimes, third-party assurance, a vocabulary of standardised metrics, and a new word for the gap between what companies claimed and what they actually did. For sustainability the word was greenwashing. For artificial intelligence the word already exists. It is AI washing, and the regulators have started using it.

The useful thing about having watched the ESG cycle is that you can see the AI cycle coming, and roughly where it has got to. So let me set out where I think it is, and what it means if you run IR at a company that has just discovered it has an AI story to tell.

An unverifiable claim is still priced, just not on its content

Start with the mechanics, because the mechanics are the whole point. A disclosure that cannot be verified is not worthless. It is priced. It is simply priced on something other than its content: the credibility of the person making it, and the cost they would bear if it turned out to be false.

This is ordinary disclosure theory, and it applied to carbon long before it applied to compute. When a company makes a soft, forward-looking claim the market cannot check, investors do one of two things. If the company has a track record of saying things that turn out to be true, and a regulator standing behind it that can punish a lie, the claim moves the price. If neither condition holds, the claim is discounted to noise. Worse, the act of making a loud unverifiable claim can become a negative signal in itself, because honest companies tend not to need to shout.

That second case is the one to worry about. A growing body of work on corporate AI disclosure finds the asymmetry you would expect: institutional investors do not treat all AI talk alike, and they are getting better at separating the firms that disclose AI as a managed risk from the firms that disclose it as a marketing flourish. The disclosure that helps you is specific, bounded, and slightly boring. The disclosure that hurts you is enthusiastic.

A fair objection at this point is that AI is not ESG. Sustainability claims were largely reputational, while AI touches the operations and the numbers directly, so the disclosure is arguably easier to ground in something real. That is true, and it cuts the right way. It means the firms with genuine substance have more to point to, and the gap between them and the performers will widen, not narrow, once the market learns where to look.

The regulators are not waiting for the standards

The reason this matters now, rather than in three years when the frameworks harden, is that enforcement has run ahead of standardisation. In the United States the SEC brought its first “AI washing” cases in 2024, settling with two investment advisers for overstating their use of artificial intelligence, without waiting for a dedicated AI disclosure rule. The general anti-fraud machinery turned out to be quite enough. If you say you use AI and you do not, the regulator does not need a new statute. It needs your marketing deck and your back end.

In Asia the posture is different, and for IR Advantage’s clients it is the more important one. Singapore’s Model AI Governance Framework, first for generative AI and now extended to agentic systems, sets out in detail what responsible AI adoption looks like, framed as guidance rather than hard law. The HKMA has mapped responsible generative-AI adoption across Hong Kong’s financial sector. The ASEAN Guide on AI Governance and Ethics has done the regional version. Hong Kong’s exchange, in its 2025 governance guide, told boards in as many words to treat the use of AI as something to be controlled and assessed, not announced.

Read those documents together and a pattern emerges. The Asian frameworks are arriving as expectations before they arrive as rules. That is both a gift and a trap. It is a gift because the company that adopts the substance early gets to define what good looks like in its market. It is a trap because “voluntary” is exactly the register in which performative disclosure flourishes. There is no filing to get wrong, only an impression to manage, and impressions are cheap to fake until suddenly they are not.

The residual in AI disclosure is judgment, not vocabulary

The part of AI disclosure that matters is the part a template cannot supply.

Any company can now generate the AI-governance section of its annual report. The frameworks are public, the language is converging, and a model will produce a creditable paragraph on human oversight and risk assessment in about four seconds. Which means that paragraph is worth roughly what the AEO ROI numbers in the last instalment were worth. It is table stakes that signal nothing, because everyone can produce it.

What signals something is the judgment underneath. Deciding which AI uses are material enough to disclose and which are noise. Deciding what you will not claim because you cannot yet stand behind it. Deciding whether your AI story is a governance story, meaning we are managing a powerful tool carefully, or a growth story, meaning this will transform our margins. The market reads those two as opposites. The first is credible because it is costly to the ego. The second is suspicious because it is free.

Hong Kong’s regulators have a phrase for this that predates the AI debate and survives it intact: substance over box-ticking. The whole of responsible AI disclosure reduces to that phrase. The substance is a human act of judgment about your own honesty. The box-ticking is what the machine will happily do for you, and what the market has already learned to ignore.

Before the market can verify your AI, it will price your disclosure of it.

So what, if you run IR

The sustainability analogy is not a comfort. It is a warning with a timetable. The companies that came out of the ESG cycle with their credibility intact were not the ones with the glossiest reports. They were the ones who disclosed early, disclosed conservatively, and never claimed a number they could not later defend in a room with an analyst who had done the work.

The AI version is the same discipline, arriving faster, with the regulators already awake. So the move is not to wait for the standard and then comply with it. It is to decide now, deliberately and in your own voice, what your company genuinely does with AI, what it does not, and what it is not yet willing to promise. Then disclose that, in that order, before the vocabulary calcifies and everyone sounds the same.

Because they will all sound the same soon enough. The frameworks guarantee it. And on the day the AI-governance paragraph becomes universal and free, what sets a company apart will be the restraint behind it: the claims it chose not to make while the choice still cost something.