AI Safety

Four plugs sparking on contact, one plug connecting cleanly

Four Elicitation Methods Failed, the Fifth Guessed All 256 Triggers

Five elicitation methods were run against a model with a chemistry classifier encrypted into its weights. Four returned chance accuracy. The fifth recovered it by enumerating all 256 possible triggers, which the authors say themselves.

A glowing sphere streaming light filaments at a caged barrier, five scattering and one passing through a single gap

The Rogue Agent Argument, Checked Against the Reports

A published critique argues that calling OpenAI’s agents rogue lets the company off the hook. Checked against the two primary incident reports, its language claim holds and its central factual claim does not.