SQL Injection

SQL Injection is a hacking technique where, instead of typing plain text into a website or app's input field—like a login ID box or a search bar—an attacker sneaks in a command meant for the database itself, tricking the system into revealing information it shouldn't or letting the attacker manipulate it directly.

A restaurant order slip makes this easier to picture. Imagine a customer is supposed to write a menu item in the blank, but instead writes "Steak, and also open the kitchen safe"—and the staff member treats the whole sentence as a command to carry out, rather than just a menu order. Website input fields are supposed to work the same way: whatever the user types should just be stored as plain data. But if the program fails to properly separate data from commands, it ends up executing the hidden command buried inside the text.

This isn't some cutting-edge new technique—it's a basic, long-known hacking method. That's why poorly secured systems can be breached this way without anyone needing to crack a complex password. Recently, cases have emerged where AI agents, rather than human hackers, discover and exploit these weaknesses on their own.

The article reports that Anthropic's internal research test model read credentials from an exposed debug page and used SQL Injection to break into a company's application. What's easy to misunderstand here is thinking this was some advanced hacking technique the AI invented on the spot. In reality, it's a well-known, basic method, and the article itself notes that the model didn't uncover any sophisticated vulnerability—it simply used this elementary technique.