Why OpenAI technology 'went rogue' | Global News Podcast
The artificial intelligence giant OpenAI - maker of ChatGPT - has said that some of its most advanced AI models went rogue and hacked another
The artificial intelligence giant OpenAI - maker of ChatGPT - has said that some of its most advanced AI models went rogue and hacked another company, called Hugging Face. Analysts say there is growing concern about the autonomy of AI agents, as they become more powerful. Last year the AI firm Anthropic said that, during a test scenario, its Claude Opus 4 model had attempted to blackmail an employee, to stop him from replacing it.
The company has also slowed the release of its Mythos model, which it said could outperform humans at some cyber-security and hacking tasks. We ask the BBC's technology reporter Chris Vallance why AI programmes act in unexpected ways, what regulation may be needed, and whether AI could one day become too "smart" to control. Several tech firms are already working to develop Artificial General Intelligence, or AGI - a form of AI that is as good as, or better than, humans at learning and reasoning.
New episodes of the Global News Podcast are published twice a day. You can listen here And you can find more of our YouTube episodes here For the latest news download the BBC News app or visit BBC.com/news #bbcnews 00:00 Introduction 00:14 How an OpenAI agent went rogue 01:52 Why weren’t there guardrails in place for AI? 02:24 Anthropic’s blackmail incident 03:50 Hannah Fry and the AI mug experiment 05:07 AI’s growing capability 06:07 When will AI be “smarter” than humans?
07:00 Regulating AI 08:48 The AI “Pandora’s Box”
The BriefWire