Why Even 'Responsible' AI Companies Can Make the Race Worse

A post that climbed to the top of LessWrong (a popular forum where AI researchers debate safety) makes a uncomfortable point: you don't have to be the reckless one to make things worse. The author, writing under the name "testingthewaters," argues that even labs that talk openly about AI danger are still pushing the whole field forward faster — just by showing up and competing.

The post lists eight concrete ways this happens. First, competing creates artificial urgency: when smart, ambitious people feel the clock ticking, they cut corners. Second, your work acts as a signpost — you reveal which techniques are promising, like reasoning models (AI that talks through problems step by step) or RLHF (training AI to match human preferences). Everyone else copies you. Third, your participation makes reckless behavior look normal: "it's a competitive market, this was bound to happen."

The sharpest point is about credibility. Once a safety-focused company is also racing to win, its warnings start to sound hollow. People ask: if this technology is so dangerous, why are you still building it? That question weakens the very voices most likely to push for caution. The race also splits public attention and government scrutiny across many players, so no single company faces real pressure to slow down.

Finally, the author warns about the mental trap of "we must win." That mindset can push even careful teams to take shortcuts they'd never have accepted a year earlier. The essay doesn't claim there's an easy fix. Its real point is simpler and harder: staying in the race has costs, and pretending otherwise — because someone else is worse — is how everyone ends up worse together.