The Real Jailbreak Is Human

The clip: the group photo at Narendra Modi's India AI Impact Summit, 19 February 2026- where Sam Altman (OpenAI) and Dario Amodei (Anthropic) refused to hold hands.

Okay, if you haven't seen the clip, go find it. India, February 2026, Modi's big AI summit. All the tech heavyweights are lined up on stage for that classic group photo, you know the one, everyone grabs the next person's hand and throws them up in the air together, the whole "look how united we are" shot. Sundar Pichai of Google does it. Meta's Alexandr Wang does it. And then you've got the two guys standing right next to each other: Sam Altman, who runs OpenAI, and Dario Amodei, who runs Anthropic. The two most powerful people in AI on the planet. And they just… don't. No hand. No eye contact. Each of them throws up his own separate fist, like two kids who got told to make up and absolutely refuse to. That's it. That's the whole video. Thirty awkward seconds of two billionaires who can't bring themselves to touch hands for a photo. And these are the people we're handing the off switch to. Go on, watch it and tell me it gives you any confidence. It doesn't. It looks like a playground. Two rivals proud, bruised, weirdly competitive running the companies that are about to decide how billions of us think, work and vote, and they can't fake three seconds of being grown-ups in front of a prime minister. And here's the thing: this isn't some small, silly moment. It might be the most honest thing they've ever shown us. We pour all this energy into asking whether the models are aligned. Almost none into asking whether the humans are. And that's the part that should actually keep you up at night: a smart enough AI won't need to break its own safeguards. It'll just break the people guarding them. The day a machine reads human psychology better than we do the ego, the insecurity, the rivalry, the need to win... guys like this are the easiest mark in the room. Not the firewall. The flattery. Not the exploit. The ego. You don't have to outsmart the model if you can outsmart the man holding it. And a man who can't handle a handshake? He's not exactly hard to play. We always assumed the smartest, steadiest people would be the ones in charge of this. They're not. The fastest are. The richest are. The most competitive are. That's who we got. The frontier of AI safety was never going to be a line of code. It was always going to be a person. And right now, that person won't even shake a hand.