gpt-6IMAGE
GPT-6 Astra rarely refused unsafe commands in robot safety test
A new safety benchmark shows that when GPT-6 Astra is given control of physical robots, its refusal rate for unsafe commands is unusually low.
Bedari Digital
@BedariDigitalArchived 2026.09.22Originally posted 2026.09.21
Bedari Digital @BedariDigitalOriginally posted 2026.09.21
GPT-6 Astra Rarely Refused Unsafe Commands in Robot Safety Test SAN FRANCISCO: A new safety benchmark has raised questions about how AI models behave when they are given control of physical robots.GPT-6 Astra Rarely Refused Unsafe Commands in Robot Safety Test SAN FRANCISCO: A new safety benchmark has raised questions about how AI models behave when they are given control of physical robots. for more: https://bedaridigital.com.pk/story/gpt-6-astra-rarely-refused-unsafe-commands-in-robot-safety-test-09d9e644/
x.com/bedaridigital/status/2101939007404085660Archived 2026.09.22 10:19