You run Trust & Safety for a platform with millions of posts a day. You hold the dial. Move
it toward leave it up and harm flourishes; move it toward take it down and legitimate voices
get silenced. There is no setting that harms no one — try them and see.
What you’re actually holding
Notice what you couldn’t do: empty both columns. Loosen the dial and real people get harassed,
scammed, and endangered. Tighten it and real people — activists, survivors, satirists, the marginalized —
get silenced, usually the ones with the least power to appeal. The “neutral” middle doesn’t escape it; it
just makes the hard calls slowly, by hand, and owns that some of them will be wrong.
That’s the shape of a lot of power: not a puzzle with a clean answer, but a tradeoff you have to own.
Wielding it well isn’t finding the magic setting — there isn’t one. It’s being honest about who bears which
harm, giving people a real way to appeal, and not hiding behind “we’re just neutral.”
And handing the dial to an AI doesn’t dissolve any of this. It makes the same two mistakes —
confidently, at a scale no human team could match, often with no one left to hear the case that doesn’t fit
the pattern. Faster and cheaper isn’t wiser. The judgment the hardest calls need is exactly what gets
automated away.