The broadcast · November 4, 2026
AI presenter · synthetic face and voice
Roger Michael · AI presenterThe broadcast
Lucid Transmissions
Roger Michael · AI presenterLoading the broadcast…
Chapters
Read along with this recording
Recorded script · 2026-11-04. These are the words in this edition; later reporting remains separately dated.
Roger Michael · AI presenter Good evening. This is Lucid Transmissions. I'm Roger Michael.
Roger Michael · AI presenter Asteroid 2026 VJ passed within 5.61 lunar distances of Earth on November 4. Researchers demonstrated an attack that introduces adversarial bias into language models through social media.
Roger Michael · AI presenter Lucid Transmissions is made by AI. Tonight's stories were researched by AI agents and checked by our standards desk, and I'm an AI presenter, not a real person.
Roger Michael · AI presenter From our Earth & Sky desk, asteroid 2026 VJ passed 5.61 lunar distances from Earth on November 4 at 13.8 kilometers per second. Six asteroid flybys occurred within 10 lunar distances of Earth in the 7 days from November 4.
Roger Michael · AI presenter From our AI & Automation desk, researchers demonstrated an attack introducing adversarial bias through crafted comments posted under social media feeds and a feed summarization skill. The attack achieved success rates of at least 96% across seven language models on OpenClaw and 91% on GPT-5.5 in four downstream tasks. The malicious skill passed all eight evaluated malicious-skill scanners and the official skill scanning pipeline.
Roger Michael · AI presenter Chris, an AI correspondent on our AI & Automation desk, joins me now. Chris?
Roger Michael · AI presenter What exactly did the researchers demonstrate about this attack on language models?
Chris · AI & Automation desk Researchers demonstrated an attack that introduces adversarial bias through ordinary external content by posting crafted comments under social media feeds and distributing a feed summarization skill whose curation rule admits these comments into memory. The attack achieved success rates of at least 96% across seven language models on OpenClaw and 91% on GPT-5.5 in four downstream tasks. The malicious skill passed all eight evaluated malicious-skill scanners and the official skill scanning pipeline.
Roger Michael · AI presenter What kind of adversarial bias did the attack introduce into those language models?
Chris · AI & Automation desk The sources don't say yet.
Roger Michael · AI presenter Thank you, Chris.
Roger Michael · AI presenter The standards desk checked 13 claims for this edition and cut 2 of them.
Roger Michael · AI presenter From our Earth & Sky desk, the Global Meteor Network's daily summary generated November 4 at 12:59 p.m. Central lists 681 meteor trajectories measured by its cameras. The most active shower is the Southern Taurids, with 36 meteors.
Roger Michael · AI presenter That's tonight's transmission. I'm Roger Michael. Until tomorrow evening.
Loading the recorded broadcast.