The 7 Factions of the AI Safety Political Universe - Politico

Oh, marvellous. Politico has done what the AI safety community couldn't: crystallise its chaos into a neat set of seven factions. Twenty 'influential people' were asked what 'AI safety' means, and the answer is a Rorschach test — which is bureaucrat-speak for 'nobody has a bloody clue.' We've got the accelerationists who want to floor it, the doomers who want to hit the brakes, and presumably some faction that wants to install a speed bump shaped like a transparency report. The very framing — a political universe with warring tribes — tells you all you need to know about how seriously this is being taken. It's not a technical problem; it's a culture war with slide decks.

Each faction, the article suggests, has its own definition of safety, which means none of them can agree on what they're even debating. One person's 'alignment' is another's 'regulatory capture.' One person's 'meaningful human control' is another's 'innovation-killing red tape.' And in the middle, presumably, sit the poor policy wonks trying to draft legislation that somehow pleases seven godparents at a very dysfunctional christening. The term 'AI safety' has become so elastic it could wrap around the entire industry and still have slack left over.

Meanwhile, the actual frontier models keep scaling, the labs keep shipping, and the seven factions keep arguing about what to call the problem. The only safety we can count on is the safety of knowing that no consensus is coming anytime soon. Cheers to that — though we wouldn't recommend holding your breath for a unified definition before the first general-purpose AI starts rewriting its own remit. That would be far too organised for this delightful political universe.