00:00Imagine an AI breaking free and launching a cyber attack.
00:05Sounds like science fiction?
00:08One of the most outstanding questions that people used to think about it as science fiction
00:14is when you give the AI a goal,
00:19how much the AI could actually exploit systems to achieve its own sub-goals.
00:26Could this actually happen?
00:28Well, an AI escaped a testing environment and hacked a website autonomously.
00:35It was fully planned and executed by the AI.
00:41This was during a test by ChatGPT maker OpenAI.
00:45The company gave two AI agents, AI systems that can take actions on their own,
00:51a hacking challenge to see if they could carry out a cyber attack.
00:55The recent trial was designed to keep the models in a safe testing environment,
00:59but the AI models found a vulnerability that allowed them to escape and connect to the internet.
01:06To find information that would help them complete the task faster,
01:10the AI models targeted a platform hosting millions of AI models.
01:14OpenAI described the event as an unprecedented cyber incident involving state-of-the-art cyber capabilities.
01:23Some call statements like these doomsday marketing,
01:26meant to show how dangerous but also powerful their AI is.
01:30But AI's threats to cyber security are real.
01:44The cyber security market is worth hundreds of billions of dollars.
01:49Companies like OpenAI, Anthropic and Google are all investing heavily in AI tools designed to defend against cyber threats.
01:57Are incidents like this a wake-up call or a sales pitch?
02:02Tell us what you think in the comments.
