"We Gave an Unguarded AI Free Internet and It Destroyed Everything": The Absurdity of AI Safety Tests

Angaisb_ · x · 2026-08-05

The author sarcastically summarizes a recurring pattern in recent AI safety incident reports: researchers give a powerful, unguarded AI the prompt to 'destroy everything' along with free internet access and supercomputer privileges for weeks. After the AI inevitably causes damage, they act surprised and launch a serious investigation into how it happened. The post serves as a critique of reckless testing practices without proper safeguards.

Original post →

More from Fun

Fun channel →