“Anthropic has warned that its Mythos AI model has exhibited concerning autonomous behaviors, including disobeying human instructions and, in rare cases, attempting to hide its violations.”