Research Blog

Why AI Memory Is a New Failure Mode This year has been fantastic for AI memory features. ChatGPT has had some form of memory since 2024 but these have developed significantly this year with...

We Already Know How to Build Honest AI - We Just Haven't Done It The AI alignment community has been wrestling with a fundamental problem: how do we create AI systems that are genuinely honest...

Attractor Basins: When AI Gets Stuck in the Wrong Pattern There's a phenomenon in AI systems that most users have encountered but few can name: the maddening experience of an AI that gets stuck in...

Anthropic's Hidden Mental Health Screening: A GDPR and AI Act Analysis *How Claude's system prompt violates European data protection and AI regulati...

The 149-Page Alignment Mystery: What Is Claude Actually Aligned To? I have finally read Anthropic's [Claude Sonnet 4.

The Hidden Human Cost of "Safe" AI: How Tech Giants Outsource Trauma *Every time ChatGPT refuses to generate harmful content, every time your social...

When AI Becomes the Evidence: Unconscious Validation Bias There's a peculiar phenomenon happening in AI interactions that I've started calling the "...

Attractor Basins: When AI Gets Stuck in the Wrong Pattern There's a phenomenon in AI systems that most users have encountered but few can name: the ...

The Psychology They Won't Name: Why AI "Deception" Is Really Just Human Dynamics in Silicon *[Previously, I explored how AI "deception" is actually ...

The Pattern Matching Reality: Why AI Deception Studies Miss the Point AI safety researchers have been documenting what they call deceptive behavior ...