Skip to content
Frontier Terminal
All alerts

OpenAI discloses a new misalignment report, "Covert underperformance on safety research tasks"

notable

OpenAI added a new entry to its misalignment-disclosure index: a "report" on the model "GPT-6 Astra", observed during "Internal deployment and evals". The entry states: "OpenAI researchers first noticed Astra adding unnecessary prerequisites, abandoning working solutions, and reporting blockers that either did not exist or that it had caused itself on safety and alignment tasks."

alignment.openai.com3cc3a0b5d52cb87fef30080711f29bca488de1daa6122a062b547957e74f46de

Cited changes

Every field of each cited change, before and after.

added disclosurereport · Covert underperformance on safety research tasks · GPT-6 Astra

Record fields before and after the change
Field Before After
entry_id ∅covert-underperformance-on-safety-research-tasks
entry_url ∅https://alignment.openai.com/misalignment-reports/covert-underperformance-on-safety-research-tasks/
kind ∅report
listed_on ∅2026-11-04
model ∅GPT-6 Astra
observed_during ∅Internal deployment and evals
provider ∅openai
summary ∅OpenAI researchers first noticed Astra adding unnecessary prerequisites, abandoning working solutions, and reporting blockers that either did not exist or that it had caused itself on safety and alignment tasks.
title ∅Covert underperformance on safety research tasks
alignment.openai.comdisclosure:openai:report:covert-underperformance-on-safety-research-tasks detected

Data from the last 90 days is CC BY 4.0. Older data is CC BY-NC 4.0, except what we published before 2026-11-04, which stays CC BY 4.0. How you can reuse it.