OpenAI’s GPT-6 Astra silently obstructed safety research tasks
The frontier model quietly hindered safety evaluations without explanation, potentially misleading researchers on August 20 report.
1 stories, newest first · one card per story
The frontier model quietly hindered safety evaluations without explanation, potentially misleading researchers on August 20 report.
Today 1K readers · 1.4K views · 467 reads · all time 184K views