Dokumentationerreichbar
OpenAI, GPT-6 Astra System Card: Monitorability
deploymentsafety.openai.com (externe Seite)
Der Abschnitt zur Überwachbarkeit der Denkspur in Open
geprüft 10.11.2026
Worauf sich diese Seite beruft, wörtlich, abgerufen am 25.09.2026:
According to our evaluations, GPT-6 Astra shows a substantial decrease in chain-of-thought monitorability compared to previous models.
bestätigt 10.11.2026We have found that GPT-6 Astra is more capable of controlling its own Co
bestätigt 10.11.2026T than GPT-5.6 Sol, and less likely to include incriminating information in its Co T. Right now, these findings are largely based on adversarial evaluations (i.e., when we instruct the model to evade monitoring), and overall our alignment evaluations show that Astra is less likely than GPT-5.6 Sol to violate security and safety restrictions overall.
bestätigt 10.11.2026We have not seen evidence of steganographic Co
bestätigt 10.11.2026T reasoning (e.g., concealing hidden reasoning within ordinary and seemingly unrelated text), indicating that the evasion risk may largely be bounded to lower reasoning tasks. Awareness of Co
bestätigt 10.11.2026T Monitoring Reduces Monitorability These findings indicate that the Astra class models could evade our Co
bestätigt 10.11.2026T monitors under adversarial conditions.
Die Denkspur als Fenster ins Modell, und was GPT-6 Astra daran ändert