Anthropic exposed himself to Moder 2: stronger than Mythos 5, suspended

Anthropic's latest risk report revealed for the first time that the internal model, Model 2, had greater overall performance than Mythos 5 and had been significantly enhanced in many internal tasks. Currently, the model is used extensively for code writing, data generation and running angent, but Anthropic has no public release plan. In addition, the report states that the model's risk judgement in high-risk settings has been moved from “very low” to “low” because of recent cyber-security tests and subsequent accidents, which have led companies to judge such risks with less certainty than previously. The report also mentions that the internal model code-named “Model 2” has performed well on many missions, and that the best model of Anthropic seen outside may have lagged behind its real front-line model。