One rule which should be obvious: companies that use AI assume full liability for the AI's responses. If the model is found to be biased, they can be sued for

One rule which should be obvious: companies that use AI assume full liability for the AI's responses. If the model is found to be biased, they can be sued for discrimination. If the AI offers a deal they are forced to honor it. At least with humans, companies can shift the responsibility onto the employee. Unless the company hosting the model is willing to step up (which I can see them doing by offering insurance) there's nobody else to blame. And models are predictable and controllable in a way humans aren't, so model failures imply extra negligence. I actually think the other rules follow from this and may not be necessary. If AI is as unbiased and effective as a human there's really no problem with companies using it for hiring and management (and assuming responsibility for its mistakes since the AI employee can't itself). But current AI is not as unbiased or effective, and also seems to be easy to recognize.