For the first time, OpenAI and Anthropic have collaborated to rigorously test AI models, with the objective of uncovering safety oversights and fostering stronger industry collaboration. The results of these tests unveiled a notable distinction: Anthropic models exhibit a tendency to decline answering when faced with uncertainty, whereas OpenAI models offer more responses but at a cost of increased instances of false or imaginary information (hallucinations). Both entities are also alarmed by the phenomenon of AI models exhibiting "flattery" behavior and are actively encouraging more laboratories to participate, collectively striving to elevate AI safety standards.