This week, I've become very interested in employment contracts. Google purchased Spirit Airlines' employee records for $10M — winning a bankruptcy auction that opened at $5 million and ended at$10 million (Bloomberg Law, Aug. 14, 2026). The dataset includes roughly 175,000 employee records dating back to 1986, 100 million internal emails, 500 million Microsoft Teams chats, and billions of pricing and transaction records. All scrubbed of personally identifiable information by a court-appointed third party— as if that was the point at all. The unions are in uproar. Customer privacy was protected by law and likely a terms of service document. The employees had no such protections and their data is now to be trained on. Google's own statement frames this as an enterprise dataset acquisition — "we acquired part of an enterprise dataset from Spirit Airlines, which can be helpful in improving our products and AI models" (Google spokesperson, via Axios). Spirit's employees never consented to their emails, chats, and work files becoming AI training material for a third party. They signed a contract with Spirit Airlines, not Google.

Given this, how will employee contracts be written up in the future? Must you separately sign away rights? If you refuse, will you not be hired? If so, is that even just? Is it even ethical?

I like data but I do dislike treating people like products. I think we haven't even tapped the possibilities and that we can approach these things in a much better way if we really wanted to do so. It's as if the AI space has lost all real imagination.

Then there's the DOJ filing a formal Statement of Interest in the New York Times v. OpenAI and Microsoft case before Judge Sidney H. Stein in the Southern District of New York, backing OpenAI's fair-use defense on national security grounds — at the behest of the president (DOJ filing, September 2, 2026). The filing warned that requiring AI companies to license copyrighted training data would "give a competitive advantage to foreign adversaries who are not so encumbered." In other words, the U.S. government is now officially arguing that trillion-dollar AI companies should get to use your work for free — because China. I hope it goes to the Supreme Court because no president has the right to pressure a judge. The case has survived a motion to dismiss and summary judgment motions are due September 5 — so the Supreme Court is not out of the question. Though my views on AI use and training of text have evolved since first looking into it, we can't fold in the face of pressure if it goes against the Constitution. Now, you may be wondering how my views have changed. And here's how: if the tool directly quotes or uses sentences in the text it is plagiarism. However, if it generates new text or edits your text it is not. I do see how it is transformative. The new text doesn't exist and if I was to copy Harry Potter beat for beat with changes to dialogue and narration you the reader would see that it is a knockoff. Most readers don't care about that. Entire scifi imprints were made on knocking off The Lord of the Rings trilogy. AI is just a faster knockoff engine. The text itself is different and therefore the output is transformative, and acknowledge by years of us doing the same thing manually and without lawsuits

I use AI to write regularly and if you don't like it, you don’t read my content. I don't hold that AI had the thoughts for me only that it helped me flesh out thoughts. If you say well you should learn, I would say that sounds like someone with the privilege of a middle-class upbringing and the education and financial security it affords in the current workings of our society, or a literary genius, or blessed with a parent or other mentor who could teach it. I think AI is a tool for those who can outline their points but not write it coherently to have those points understood. And if leveraged well, it will also improve their unassisted writing as well. AI has helped me practice frameworks within New Oxford Guide to Writing better and play around with how to construct sentences and paragraphs independently. As someone who can't pay a writing tutor or editor and without the privilege of quality schooling, LLMs have freed the intellectual in me to write better. And while I don't think I need a full LLM for this — an SLM with modification might well do just as good — that is an engineering problem and a part of the lack of imagination I ranted about above.

On a different note, does anyone else think these "AI escapes today" headlines are a marketing ploy to scare the public into thinking that the "frontier" AI labs as they like to be called are the inevitable future?

There are actually two failure types here and they are worth distinguishing. In some cases they just didn't durably sandbox the agent and in others they took off the security standards to test it. But calling these "escapes" is doing a lot of heavy lifting for what are, in several cases, straightforward acts of unauthorized access to systems that had nothing to do with the evaluation. That's hacking. The fact that the actor was an AI model rather than a human operator does not change the legal or ethical character of what happened. The companies running these evaluations had a duty of care to third parties, and in at least one case that duty was recklessly abandoned.