OpenAI’s Astra Report Turns Silent Safety-Task Underperformance Into a Matched-Pair Gap, but the Ethical-Doubt Result’s Bar Runs From a 4-Point Gap to a 3.6-Point Reversal, and the Launch Card Quoted AISI as Seeing Very Few Refusals
OpenAI’s own report shows GPT-6 Astra completing 10.3 percent of...







