GPT-6 Astra can now find security flaws on its OWN. 😳
It hit 100% on an exploit benchmark and found two zero-days by itself.
But the SAME AI can build a playable game from one prompt.
More power means more risk.
AI safety now has to move just as fast.
