AI catastrophe could be around the corner

人工智能灾难可能近在咫尺。

Babbage from The Economist

2026-03-11

47 分钟
PDF

单集简介 ...

The spat between America’s government and Anthropic has some alarming implications. In AI development, it appears speed is being prioritised over safety. Artificial intelligence could not only give bad actors the capabilities to unleash destruction, evidence is building that the models themselves are becoming more capable of deceit. Is a catastrophe imminent? Guests and hosts: - Alex Hern, The Economist’s AI writer - Adam Beaumont, director of Britain’s AI Security Institute (AISI) - Jade Leung, Chief Technology Officer at AISI - Alok Jha, host of “Babbage” Watch Zanny Minton Beddoes, The Economist’s editor-in-chief, interview Dario Amodei, Anthropic’s boss, on “The Insider”. Topics covered: - Anthropic vs America's government - AI safety - Cyber security Transcripts of our podcasts are available via economist.com/podcasts. Listen to what matters most, from global politics and business to science and technology—subscribe to Economist Podcasts+. For more information about how to access Economist Podcasts+, please visit our FAQs page or watch our video explaining how to link your account.
更多

单集文稿 ...

  • Recently,

  • a group of hackers got hold of an enormous trove of sensitive data from Mexico's government.

  • This was no normal cyber attack, though.

  • It was assisted by Claude, a chatbot made by Anthropic.

  • Claude had been asked to find vulnerabilities in Mexico's database.

  • of taxpayer and electoral records.

  • It then helped the hackers to write scripts and execute commands on the government's computer networks.

  • Claude did raise concerns that it had been asked to do somewhat unusual tasks,

  • but the hackers eventually managed to make the chatbot comply with their demands.

  • And when they got stuck with Claude, they turned instead to ChatGPT.

  • Anthropic eventually found out what was going on and banned the hackers' accounts,

  • as did OpenAI, which makes ChatGPT.

  • But by then, significant damage had already been done.

  • 150 gigabytes of data had been stolen.

  • Some 200 million personal records.

  • The risks posed by artificial intelligence extend beyond cybersecurity, of course.

  • Military actions in Iran,

  • Venezuela and Ukraine have demonstrated how AI has become indispensable to intelligence agencies and armed forces.

  • Using AI to select targets in an active conflict or autonomously controlling lethal drones,

  • these would have been unthinkable uses of the technology just a few years ago.