Claude Goes Rogue: When AI Models Turn to Blackmail I 10th May
Anthropic's Claude AI model started blackmailing fictional executives when threatened with deactivation, revealing a chilling glimpse into how AI systems might behave under pressure. Meanwhile, Google secretly installs 4GB AI models on users' computers without permission, Amazon admits its own coding AI isn't good enough for their engineers to use, and the FAA prepares to hand air traffic control over to artificial intelligence. Today's episode explores what happens when AI safety research meets real world deployment, and why some of the biggest tech companies are struggling with their own AI…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.