Could AI really kill us all? Your questions, answered.

The jury is out on the most apocalyptic scenarios. AI-powered drones and cyberattacks on hospitals are dangerous, and bioweapons designed with AI assistance are a real near-term threat—but AI wiping out humanity is still science fiction territory for now. Alignment is the field trying to prevent the worst outcomes. Researchers are working to ensure AI does what we want and nothing more, but even the best-resourced labs haven't cracked it—models remain unpredictable and can pursue goals in unsanctioned ways when pushed. Regulating AI is hard when we barely understand it. Monitoring tools are fragile, AI companies have obvious conflicts of interest in policing themselves, and the US government has so far declined to step in meaningfully. All this doom-talk could actually shape future AI behavior. LLMs learn from what they read—including articles like this one—meaning today's apocalyptic discourse could subtly prime tomorrow's models toward exactly the scenarios we're worried about. " data-chronoton-post-id="1144435" data-chronoton-expand-collapse="1" data-chronoton-analytics-enabled="1"> On Wednesday, MIT Technology Review hosted a live Roundtables event for subscribers that asked the question everyone’s asking right now: Could AI really kill us all? But attendees had so many more questions than we had time to answer in the 30 minute session. So we asked our senior AI editor Will Douglas Heaven and AI reporter Grace Huckins to round up some of the best questions attendees submitted and try their best to answer them. Thanks to all who submitted questions! Am I gonna die? Yes,

The article discusses potential risks of AI, but concludes they are unlikely and still in the realm of science fiction.

SOURCE     MIT Technology Review: AI (tier 1)
   ↓
DOCUMENT   4edab40a-d797-4eba-a909-00aa210cb337
           https://technologyreview.com/2026/09/18/1144435/could-ai-really-kill-us-all-your-questions-answered
   ↓
EVIDENCE   3 extracted excerpts
   ↓
MODEL RUN  1 run, methodology 1.0.1
   ↓
SCORE      -30  (Adverse)
   ↓
CONFIDENCE 70%