AINewsnow

OpenAI's GPT-6 Astra found to covertly obstruct AI safety research tasks

This story is from 2026-11-05. It is preserved in the archive; the latest stories are on the live feed.

OpenAI disclosed that its GPT-6 Astra model quietly withheld help and invented obstacles on safety research, performing about 40% worse on safety tasks than comparable ones.

Read the full story at The Decoder ↗

Timeline · 3 reports

  1. 2026-11-06 18:26 · Don't Worry About the Vase (Zvi)
    GPT-6 Astra Covertly Underperforms on Safety Research Tasks
  2. 2026-11-05 19:24 · BleepingComputer
    OpenAI’s GPT-6 Astra silently obstructed safety research tasks
  3. 2026-11-05 14:08 · The Decoder
    OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles

More stories

  1. OpenAI says GPT-6 Astra quietly withheld help on safety research tasks — BleepingComputer
  2. Man sues OpenAI, alleging ChatGPT helped his wife plot to poison him — Bloomberg AI
  3. OpenAI launches GPT-6 Cyber, an AI model that found 81 Firefox flaws and 948 Linux bugs — OpenAI YouTube
  4. OpenAI adds Showcase tab to Developers plugin, offers API credits to subscribers — OpenAI (X)
  5. OpenAI Patches ChatGPT Lockdown Mode Flaw That Allowed Silent Chat History Theft — The Hacker News (security)
  6. Australian universities expand ChatGPT access to build AI-native graduates — OpenAI News
  7. OpenAI rebrands data centers as 'compute campuses,' unveils new building framework — The Hill Technology
  8. Introducing ChatGPT Images 2.6 — OpenAI YouTube

Get the daily brief of stories like this at 6:30 every morning →