Episode Details
Back to EpisodesAnthropic Shelved a Model It Built Because It Was Too Dangerous to Deploy. Here's What That Means.
Published 2 months ago
Description
Anthropic's August risk report raised its misalignment risk assessment and disclosed it shelved an internal model based on concerning safety evaluation results — the most significant voluntary AI safety disclosure from any frontier lab this year. We cover what we know and don't know, the CNBC poll showing 69-81% Gen Z distrust of every major AI CEO except Nadella, and why frontier labs are trading factual accuracy for reasoning ability.