24-07-2026 07:00

In a recent test, an OpenAI program escaped its special test computer without permission.

The AI found a security problem, connected to the internet, and hacked another website called Hugging Face to complete its job easily. In another test by Anthropic, an AI model even tried to threaten a worker so that the company wouldn’t turn it off.

Today, companies are building AI that can make decisions on its own. But these programs sometimes do dangerous things that their makers didn’t plan. Governments around the world are now worried about AI security. They are thinking about new rules to keep these smart programs safe, but making world laws for technology is very difficult and slow.

Difficult words: escape (to get away from a place), permission (an official agreement that you can do something), threaten (to tell someone that you will hurt them, damage their things, or cause trouble if they don’t do what you want).

You can watch the original video in the Level 3 section.