GPT-6 Astra appears to leak neuralese internal thinking traces openly

ivan_bezdomny · x · 2026-09-06

A user observed that GPT-6 Astra seems to leak its internal thinking traces (neuralese) with no effort to hide them. An example shows the model emitting internal notes like "saw new packet, only 9 groups. Do NOT stop or force fake 100. Keep those 9 fully new groups, then expand to 100 pairs." The raw internal instruction exposure raises questions about interpretability and output integrity.

Related event: GPT-6 Astra reportedly leaks unfiltered neuralese traces(2 posts)→

Original post →

More from Models

Models channel →