Essay warns AI reasoning may become invisible to monitoring

What the research found

Speculative essay reporting three OpenAI safety researchers dismissed in Oct 2026 and a reported decline in CoT monitorability in GPT-6 Astra's Sept 2026 evaluation; causal link to a loop-type Transformer is unproven, so background only and overlaps #4015.

Claims and how well the evidence supports them

1. Three OpenAI safety researchers were dismissed in October 2026 96%
2. GPT-6 Astra September 2026 evaluation showed decline in CoT monitorability 88%
3. OpenAI next model adopts a loop-type Transformer 84%

Evidence

• note.com https://note.com/emilia_lab/n/n22c1fe5aab36?hl=en

Actionability rated low.