Quite! It's worse than that - the AI companies all release tooling that deliberately gives AIs as many capabilities (arbitrary code execution) as possible to be

Quite! It's worse than that - the AI companies all release tooling that deliberately gives AIs as many capabilities (arbitrary code execution) as possible to be really sure it is not box contained. And the market rewards that behaviour as it makes the products more powerful more easily, despite the risks.