I also like this idea for another reason: A reader of the code knows immediately if the trait is object safe. Otherwise, they'd have to create a trait object and see if it fails to find out.
However, I think the syntax is a bit confusing. My suggestion is to use an #[object_safe] attribute instead; this expresses clearly what the feature does. It also doesn't require changing the syntax, which is already too complicated in many people's eyes.