Or...
Why Artificial Intelligence Is No Different Than a Good-Sized Rock
I read an article positioned as a red-alert warning that the new ChatGPT-5 model was “nearly unusable” in an enterprise environment.
The premise?
Two firms “jailbroke” the model, one through “advanced storytelling” techniques and the other through “advanced obfuscation.”
One got it to explain how to make a Molotov cocktail. The other, a bomb.
Upon reading this, I immediately began planning to upgrade my door locks and stockpiling toilet paper. But then I remembered something important:
Anyone paying attention in high school chemistry and physics already learned this “dangerous” information. We just removed the context.
Storytelling and obfuscation? These are manipulation strategies as old as humanity. People have always used charm, distraction, or clever wording to get what they want. From buttering up your boss before vacation requests to knowing exactly where you can speed on the drive home without getting caught.
And just so we’re clear, this article waved the “enterprise” flag without ever marching in that parade. From a corporate security perspective, if someone inside your company is spending 24 hours using advanced manipulation techniques to extract Molotov cocktail recipes from your AI, you don’t have an AI problem. You have an HR problem. And probably a security problem. And definitely bigger problems than jailbreaks.
Now let’s talk about what actually matters: our relationship with tools and technology. Here’s an uncomfortable truth — the difference between a tool and a weapon is the person holding it.
Take the good-sized rock from the title and you can wrap it in mortar to build a wall. Strike metal with it to shape a tool. Use it to bash your neighbor because their cave is bigger and you feel like movin’ on up.
The same goes for modern tools. A database can store decades of cancer research or millions of stolen credit cards. A surveillance camera can alert you to an intruder at 3 a.m., or spy into the neighbor’s bedroom. A pencil can write down wisdom that lasts centuries, or… well, we all saw John Wick.
A person determined to make a Molotov cocktail will figure it out. They can search the internet, ask Reddit, find a private Discord server, or simply pay attention in Chemistry class. Spending 24 hours manipulating an AI is probably the least efficient option.
The inconvenient truth is that how they get the information isn’t the real problem. The intent behind seeking it is. We are the real issue.
Maybe instead of pouring billions into engineering tools that protect us from ourselves, we invest in understanding why one person sees a rock and another sees a weapon and how to change that.
That research would be massive: sociology, psychology, mental health, environmental factors, identity, millions of data points. I get why it’s daunting. But if we’re serious about “AI safety,” maybe it’s time to stop pretending the danger starts with the tool.
If only there was some really advanced technology that could reason with this much data. Oh, wait.............