The AI model that’s ‘too powerful’ to be released to the public

单集文稿 ...

• 00:00

  This BBC podcast is supported by ads outside the UK.

• 00:05

  Every sunday we talk about the week's tech news on this week in tech hi this is leo laporte inviting you to join me this week

• 00:12

  with lisa schmeiser dan patterson and yanko retkers we 're going to talk about the new 49 megabyte web page

• 00:19

  it 's the standard you know we 'll also talk about elon musk you 've got some spleen to do and the yassify filter new

• 00:27

  from nvidia that 's this week on this week in tech you 'll find it at twit.

• 00:30

  tv or wherever you get your podcasts Most of us have interacted with an AI chatbot for something by now.

• 00:41

  ChatGPT, Grok, Gemini, there are many.

• 00:45

  But recently, Anthropic, a leading AI company and parent of the chatbot Claude,

• 00:51

  announced they 've created a model that they say is too dangerous to be released to the public.

• 00:58

  Obviously, capabilities in a model like this could do harm if in the wrong hands.

• 01:03

  And so we won't be releasing this model widely.

• 01:06

  Anthropics say that Claude Mythos Preview is frighteningly good at hacking.

• 01:12

  So good, in fact, that the likes of us can't be trusted to play with it.

• 01:16

  Banks, including Goldman Sachs and JP Morgan, are warning of the risks.

• 01:23

  And Anthropics' own researchers say they noticed Mythos was capable of being sneaky,

• 01:28

  defying instructions and covering its tracks.

• 01:34

  Has the moment arrived when we should all be terrified of an autonomous, sneaky hacking machine?

• 01:40

  Or is this all a marketing trick from an industry built on hype and bluster?

• 01:47

  I'm Tristan Redman in London.

• 01:48

  And today on The Global Story, just how dangerous are AI models becoming?

• 全部文稿