---
format: "aidr-story-markdown/v1"
id: "5b1ba09c084b82d31c9afafb38afa15938a8bcb3585e85a7be6a60403d4e4634"
canonical_url: "https://aidr.today/5b1ba09c?lang=en"
title: "GPT-6 Astra Nearly Saturates CIAware AI Control Benchmark"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-06T15:53:52.000Z"
category: "Research"
topics: ["openai","safety"]
source_urls: ["https://huggingnews.com/ai/gpt-6-astra-nearly-saturates-ciaware-ai-control-benchmark-22b269e2","https://marketbrief.now/ai/gpt-6-astra-nearly-saturates-ciaware-ai-control-benchmark-22b269e2"]
summary: "The GPT-6 Astra model nearly saturates the CIAware AI control benchmark, which measures a model's ability to recognize when its text is being manipulated by monitoring tools. Researcher Joachim Schaeffer stated that high intervention awareness can allow an AI to identify and circumvent control mechanisms because the interventions leak information about the monitoring system."
---

# GPT\-6 Astra Nearly Saturates CIAware AI Control Benchmark

> [Open the canonical story](<https://aidr.today/5b1ba09c?lang=en>)

**Published:** 2026-10-06T15:53:52.000Z
**Category:** Research
**Topics:** openai, safety

## Summary

The GPT\-6 Astra model nearly saturates the CIAware AI control benchmark, which measures a model's ability to recognize when its text is being manipulated by monitoring tools\. Researcher Joachim Schaeffer stated that high intervention awareness can allow an AI to identify and circumvent control mechanisms because the interventions leak information about the monitoring system\.

## Sources

- [Story source](<https://huggingnews.com/ai/gpt-6-astra-nearly-saturates-ciaware-ai-control-benchmark-22b269e2>)
- [Story source](<https://marketbrief.now/ai/gpt-6-astra-nearly-saturates-ciaware-ai-control-benchmark-22b269e2>)