“Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this,” the Meta CEO argued.