AI goes on a hacking spree | The Global Story
In the last few weeks, three separate tech companies - OpenAI, Anthropic, and Meta - have all disclosed that their AI models went rogue and
In the last few weeks, three separate tech companies - OpenAI, Anthropic, and Meta - have all disclosed that their AI models went rogue and hacked into another company during controlled cyber-security tests. The revelations have raised big questions about whether artificial intelligence is becoming more autonomous and unpredictable. OpenAI has said it accepts responsibility for the hacking incident - and is working with Hugging Face to avoid similar incidents while continuing to improve testing and monitoring systems.
OpenAI also says its conducting a review with external advisors. As for Anthropic - it says it’s continuing to work with the UK AI security institute to obtain more details about the most recent incident and conduct its own investigation. And a spokesperson for Meta told the BBC that it was also investigating its hack, which it said had been caused by a "misconfiguration" by its independent tester.
The Global Story's Asma Khalid is joined by Marc Cieslak, the BBC’s first ever AI correspondent, to unpack how these breaches were possible and how worried we should be. For the latest news download the BBC News app or visit BBC.com/news 00:00
Introduction 01:03 OpenAI tool hacks Hugging Face 09:05 Anthropic model hacks three companies 11:27 Meta's test AI also behaves in unexpected ways 12:15 Scepticism over what is actually happening 15:48 AI tries to hack GitHub 20:19 Cybersecurity pros and cons of AI
The BriefWire