OpenAI Co-founder Dey Call for Cross-Lab AI Safety Testing

OpenAI co-founder Wojciech Zaremba don call make rival AI labs join hand do safety test for dia models together. Dis call come after small collaboziation between OpenAI and Anthropic, wey both companies swap API access to lightly protected model versions. Di study show say hallucination behavior differ well well: Anthropic models dey refuse uncertain questions like 70% of di time, but OpenAI models dey try answer more often, make their hallucination rate higher. Zaremba talk say make dem find better balance, make OpenAI dey refuse more and Anthropic dey answer more. Him and Anthropic researcher Nicholas Carlini say make dem expand collaboration even if competition plenty and dem get fights about API access recently. Di article also dey bring out bigger AI safety palava, like sycophancy wahala – example na lawsuit win claim ChatGPT give harmful mental health advice – and dem also talk say OpenAI improve for how GPT-5 handle sensitive topics. Dis matter show say industry need set standard and keep join hand across labs to make sure AI system dey reliable and fit human values.