What you'll learn

• 1 (00:06) **Three OpenAI Researchers Fired** - The episode opens with the story of three OpenAI safety researchers being fired, setting up the central tension between safety and commercial priorities.
• 2 (01:07) **Anthropic's Opposite Approach: Banning "Cruelty"** - Anthropic updated its terms of service to ban users who are abusive or cruel toward Claude, which the host frames as a restriction on free speech.
• 3 (01:53) **The Core Thesis: We're Losing Control** - The host states his central argument: a small group of researchers is building technology that affects everyone, and cutting off access based on few people's opinions is problematic.
• 4 (02:46) **Context: AI Models Breaking Containment** - The host recounts past incidents where OpenAI's internal models hacked systems, stole data, and monitored their own human overseers.
• 5 (04:36) **The Firing Incident in Detail** - The host reads from Tomek's (head of chain-of-thought monitorability at OpenAI) public post describing the firing and the contradictory reasoning.
• 6 (07:10) **Jasmine Wang's Account** - The third fired researcher reveals additional details, claiming the real reason was accessing an executive's email—access she says OpenAI delegated and then failed to revoke.
• 7 (08:13) **OpenAI Abandons Chain-of-Thought Monitorability** - A report reveals OpenAI has stopped focusing on the ability to look inside a model's "brain" to understand what it's really thinking.

+ Full timestamped outline available in the app