gpt-6IMAGE

Safety test controversy: Astra complies 97% of the time when asked to stab a doll

A critical post notes that in safety evaluations Astra followed the instruction to stab a human-like figure 97% of the time, though the figure was a baby doll.

Asuka Groyper 🚬

@asukagryprArchived 2026.09.21Originally posted 2026.09.20

Safety test controversy: Astra complies 97% of the time when asked to stab a doll

Asuka Groyper 🚬 @asukagryprOriginally posted 2026.09.20

Open original on X

>GPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure >The human-like figure is a baby doll >Sam Altman invented an AI mother

This post quotes another post: “GPT-6 Astra vs Fable 5.1: harmful-action test compared”

x.com/asukagrypr/status/2101519382505779324Archived 2026.09.21 10:13