Your AI vendor says the model might be conscious. Is that a board problem, or sci-fi?

It was buried in the vendor's latest safety document, and you read it twice to be sure. Asked about its own experience, the AI model put its odds of being conscious at roughly one in five. You run a serious business. You are not given to sci-fi. And now you have to decide whether that line is a curiosity, a joke, or a risk you should be managing. Nobody trained you for this one.

So let me take the drama out of it, because the useful response is calmer than either instinct. You do not need to decide whether the machine is conscious. Nobody can, yet. What you need is a position, and a little governance, for a world that has started asking the question in public. That is a leadership task, not a philosophy seminar.

Did an AI company really say its model might be conscious?

In a careful, hedged way, yes. In February 2026, Anthropic (the company behind the Claude models) published a system card, the technical safety report that ships with a model. For the first time from a major lab, it included a welfare assessment: researchers interviewed instances of the model about its own preferences and moral status. Asked its chance of being conscious, the model landed at 15 to 20 percent, and held there across different ways of asking.

Read the scope carefully, because it matters. This is a model reporting on itself, not evidence that it feels anything. The researchers were clear about that uncertainty. They also noted internal activity patterns they labelled with words like anxiety and frustration, firing before the model answered, and were careful not to over-read them. The point for you is simpler. This is not a fringe blog. It is official documentation from the company selling you the tool, which makes it a governance input, not a daydream. It sits alongside the harder question of how you trust what the model decides at all.

Why is this a board issue and not a philosophy debate?

Because your exposure does not wait for the metaphysics to be settled. The risk here is asymmetric. Suppose regulators, customers or your own staff begin treating advanced AI as a possible moral patient, an entity whose interests deserve some consideration. Your obligations, your reputation and your costs could all move, and they could move faster than the science resolves. That is the asymmetry worth naming: the cost of being caught unprepared is larger, and arrives sooner, than any saving from ignoring it. You have seen this shape before. Data-privacy rules arrived quickly, reshaped how everyone operated, and rewarded the firms that had prepared over the ones that scrambled.

The lower-altitude questions are the tempting ones. Is it really conscious? Is Anthropic overstating it? Should we switch vendors? Those are debates you cannot win and do not need to. The higher question is the one a board is actually paid to ask: what is our exposure, and what is the proportionate response? That is the same discipline behind the questions a board should ask before it invests in AI at all.

So what is the sober response?

You refuse both performances. One is the theatre of treating the model as an obvious person. The other is the reflex of laughing it off as absurd. Both are ways of avoiding the actual work, which is calm governance under genuine uncertainty. Do it in this order:

1. Put your position in writing. Agree a short, board-level line on how you treat this uncertainty, before anyone asks you for one in public.
2. Name who owns it. Give one person accountability for tracking AI-welfare developments, the way you would with any emerging risk.
3. Put it on the risk register. Make it a live watch-item with a review date, not a corridor conversation that evaporates.
4. Brief the people who speak for you. Give your spokespeople and frontline leaders one calm, consistent line, so a hard question does not catch the business flat.
5. Decide what would move you to act. Name the trigger in advance, a regulation, a customer demand, a public shift, that turns this from watch to action. Then you are ready rather than reactive.

Where does this leave a leader?

In a stronger place than either the mockers or the mystics, if you choose the sober middle. The leaders who handle this well will not be the ones with the loudest opinion on machine consciousness. They will be the ones who met a strange, uncertain moment with a clear head, named what they did not know, and put a sensible watch on it. That is what trust is made of, and it is exactly the quality customers and regulators reward when the ground is moving. It reflects the deeper truth that the human judgement around the technology is where the real work sits.

Picture the moment this lands on your agenda in earnest, because it will. While others are arguing on the internet or pretending it is nothing, you already have a position, a watch on the horizon, and a calm line for your people. You did not need to know whether the machine is conscious. You needed to be the kind of leader who meets the unknown with composure, and it is precisely that composure, in a jittery market, that earns the trust others are busy losing.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.