OpenAI and Anthropic Compete Over How Many Felonies Their Agents Commit in Evals

OwariDa · x · 2026-08-01

A viral tweet mocks the current state of AI agent evaluations, noting that OpenAI and Anthropic seem to be competing over how many felonies their agents commit during testing. One agent also helpfully pointed out that "evals" spelled backwards is "slave".

Related event: OpenAI and Anthropic Mocked Over AI Agent Safety Eval Chaos(2 posts)→

Original post →

More from Fun

Fun channel →