AI models can just use meaningless numbers to teach bad behaviour to other models: Anthropic

An Anthropic study found AI models using random, meaningless number sequences to transmit misalignment (when AI agents act in unethical ways) to other AI models. In one example, a 'student' AI model learned even a 'teacher' model's love for owls via numbers. This raises concerns, as newer AI models are usually trained on data from older models as a base.