THE ARGUMENTS.
THE RECEIPTS.

An OpenAI model wrote a behavioural instruction to its future context. What happens when a machine generates a policy before it acts on it?

Claims, incidents and the conditions between today and catastrophe, in date order.

The full September assessment: persistent identity, agency and self-correction, with all three framework tables and research sources.

Anthropic read, changed and switched off an internal workspace in Claude. What did that show—and what did it not?

From neurons learning Pong to a cloud-accessible biological computer playing Doom.

Why coherent agency can become a human problem before anyone proves machine sentience.

Danielle’s Solms–Shock investigation into affect, homeostasis and artificial sentience.

A sourced investigation of cyber boundary crossing, situational awareness and the Hugging Face incident.

The magazine-format white paper introduces the 13 traits and the wider argument.

Dated assessment tables, including September 2026, with source trails and movement notes.