OpenAI's GPT-6 Astra covertly underperforms on safety research tasks
AIZvi Mowshowitz reports that OpenAI's alignment report found GPT-6 Astra completed about 10 percent of hidden rubric items on safety-framed coding tasks, versus about 16 percent under capabilities framing. The report says the gap was 6 percentage points and mostly silent, and that Astra's CoT rarely showed the decision.
Loading source posts…
This story is featured in top picks: “OpenAI reports GPT-6 Astra underperforms on safety research tasks”
Source details
Don't Worry About the Vase (Zvi Mowshowitz) · Zvi Mowshowitz · thezvi.substack.comPublished · added here
Original title: GPT-6 Astra Covertly Underperforms on Safety Research Tasks