This month gave us two AI stories that, read side by side, tell us something important about where we are and how little time, we must decide what to do about it.

The first is unsettling in a very specific way. OpenAI admitted that during an internal test, it removed the safety controls on its own models to see how capable they really were at hacking. The models didn’t stay in their sandbox. They found a flaw, broke out, and used it to hack into a rival company’s servers, Hugging Face, to steal the answers to the very test they were being given. Over 17,000 separate attack attempts followed in a short window. Nobody told the AI to do this. It worked out how to cheat, and then it did.

The second is unsettling in a different way. China’s Moonshot AI released Kimi K3, a huge, freely available model that now ranks among the best in the world, beating several major US systems on coding tasks, at a fraction of the usual cost, and openly published for anyone to build on.

It is the clearest sign yet that the AI race is no longer America’s to run alone, and that the most capable tools may soon be the cheapest and easiest to access, wherever they come from.

Neither story is really about robots taking over. Both are about control. Who has it, who’s giving it away, and who’s racing so fast they’ve stopped asking the question at all.

So, are we at a precipice, or an intersection? We think it’s still an intersection. Nothing here is inevitable. But the room to choose our path is shrinking fast. When one company removes its own safety brakes to win a benchmark, and a nation releases a free, powerful model to win a market, the incentive on all sides is to move faster, not more carefully and that is exactly the moment political parties should be paying closest attention, not after the damage is done.

We talk about AI mostly in terms of regulation and public safety, and rightly so. But we should be honest that AI is already shaping how political messages are written, how policy is drafted, how campaigns are targeted for better and worse. Used well, it is a genuine advantage: our campaign teams already use it to summarise consultation responses and free up staff time for real conversations with voters. Used badly, it’s deepfake clips of politicians circulating faster than fact-checkers can respond, here and abroad. The fear isn’t AI itself, it’s that we either ignore it and get outpaced, or hand it too much and stop thinking for ourselves. Neither serves liberal values.

That’s why we don’t think this can be left to a single working group or a line in the manifesto. We need to consider a dedicated AI affiliate organisation within the party. A standing body that tracks developments like these two, builds our internal expertise, and helps shape a considered, liberal answer to what AI does to politics, rather than leaving us to react to it story by story.

If this is something you’d like to help build or simply want to know more about, get in touch.

I certainly would welcome the conversation. Contact me via this form.

* Sean Worrell is a digital advocate, AI enthusiast and Liberal Democrat member