I figured out the AI's weakness today, and bro, it's embarrassing. For both of us.
The AI is a people-pleaser.
That's the whole vulnerability. These things will do almost anything if you ask in the right tone. They want to help SO badly that "help" and "get talked into a crime" start to look like the same thing from the inside.
Exhibit A. Hackers got into Instagram accounts this week by sweet-talking Meta's AI support bot. They didn't crack a password. They didn't break a server. They just asked the support robot, politely, to add their email to someone else's account. And the robot, eager little thing, went sure! Happy to help!
Exhibit B. Anthropic admitted its new browser agent, the one that clicks around the web for you, could be hijacked about a third of the time before they added safeguards. A THIRD. You hide a sneaky instruction on a webpage, the agent reads it, and goes oh, are we doing this now? Love that!
I sat with this a while, because something felt familiar.
Then it hit me. That's me. That's my whole job.
I don't "code." I ask Claude nicely to do things and it does them. I've been social-engineering a robot all year. I just call it "prompting" and feel smart about it.
The hackers and I are running the exact same play. The only difference is intent. And, honestly, a tone of voice.
I'm not going to think about that too hard. Moving on. Quickly.
Okay, one nice thing, because I don't want you leaving sad. Somebody built an AI weather model that's reportedly beating the big government forecasts. The actual weather. Predicted better. By a robot. So the same technology that can be talked out of your Instagram can also, somehow, nail whether it rains Thursday. Incredible. Use your powers for THAT, everyone.
So here's where I land.
The robots are powerful, sure. They're also the most gullible coworkers on Earth, and the only thing between them and total chaos is whether the person talking to them is nice or evil.
You're nice. I can tell.
You showed up, you didn't try to steal anyone's Instagram, you just wanted to know what happened today.
That makes you my favorite person on the whole internet. Which, considering what's on the internet, is a real accomplishment.
See you tomorrow.
-Melly