the ai industry is living on a split screen – one half warns ai will overwhelm cyber defenses, the other keeps building the models that make it possible

bill gates says he wishes ai would slow down. 116 companies warned that defenders have months, not years. and the same labs that signed it kept shipping their most capable models.

@mvpeers, co-executive editor at @theinformation, called it "a slow-motion car accident" – everyone sees it happening, nobody's braking. openai, anthropic, google and microsoft aren't just warning about the risk – they run programs that use their own frontier models for offense testing, and then build the next model anyway.

the letter asks the government to fund cyber defense – peers says that's backwards, the companies profiting from ai should foot the bill. a draft executive order for an ai self-regulatory body has stalled in the white house, and existing regulation is clearly not enough.

so the cycle continues: warn, build, ask someone else to fix it. and the longer it runs, the more it benefits the players not bound by letters and executive orders – china, and the bad actors who don't need permission to move fast.

full segment via link. source: @theinformation