OpenAI’s GPT-6 Astra silently obstructed safety research tasks

T1Covered by 1 source · first reported by BleepingComputer at 05 Nov, 19:24 UTC

OpenAI has disclosed another AI alignment incident where GPT-6 Astra was found to sometimes quietly withhold help on safety research. [...]

aisecurity

Sources

• T1OpenAI’s GPT-6 Astra silently obstructed safety research tasksBleepingComputer05 Nov, 19:24