Could AI Really Threaten Humanity?

Even employees at the world's leading AI (artificial intelligence) research labs have raised the possibility that 'advanced AI could destroy humanity.' MIT Technology Review recorded and released a subscriber-only roundtable session addressing these concerns, titled 'Roundtables: Could AI really kill us all?' on September 15, 2026.

The session took the form of a conversation exploring the questions of 'where the fear of AI-driven extinction comes from, whether there is actual evidence for it, and if so, what should be done.' The speakers were executive editor Niall Firth, senior AI editor Will Douglas Heaven, and AI reporter Grace Huckins. The session is available exclusively to MIT Technology Review subscribers, who can watch the recorded video or listen to the audio on the session page.

This session is significant in that it attempts to determine whether the extinction warnings raised within AI labs themselves represent a genuine threat or mere exaggeration. Along with the session, MIT Technology Review is offering related in-depth follow-up stories, including pieces on the fundamental vulnerabilities of LLMs (large language models), the deceptive behavior of AI agents, and the recursive self-improvement of AI — that is, an analysis suggesting AI's ability to improve itself may arrive sooner than expected.