Four stories worth your coffee this Friday. A humanist in Kentucky got out of using AI at work, Mistral just handed out a trillion-parameter model, Microsoft finally showed how much money OpenAI sends its way, and an OpenAI-backed holding company is buying your neighbor's alarm system.
01 of 04
A Humanist Just Got a Religious Exemption From AI at Work
David Durman is a research administrator at the University of Kentucky. His department told staff to start working AI tools into their daily workflow, and Durman objected to the ones on the list: Microsoft Teams Copilot, ChatGPT, and Claude.
He is not religious in the church sense. He is a humanist. He reached out to the American Humanist Association, then filed a request with the university's Office of Equal Opportunity under Title VII, the federal law that requires employers to accommodate sincere religious beliefs. His argument was that handing his thinking to a chatbot cuts against a core humanist commitment to thinking for yourself.
The OEO granted it in September. The university found the request did not create an undue hardship, which is the bar set by the Supreme Court in Groff v. DeJoy back in 2023. Durman does not have to personally use AI in his work, though he still has to attend meetings and submit materials through platforms that have AI baked in. No lawsuit was ever filed. It is not the first one either: a Unitarian Universalist named Erin Maus got an AI exemption in May.
Here is where I land on it. I build with these tools every day and I think most people should be learning them, not dodging them. But I like this outcome anyway. Title VII has always covered sincere non-theistic belief, and a school that decides it can live without one admin typing into Copilot is being reasonable. The bigger signal is for every company writing an "everybody uses AI now" mandate. You just picked up a new HR process whether you wanted one or not.
Sources: FindLaw, TechCrunch
02 of 04
Mistral Gives Away a Trillion-Parameter Model for Cyber Defense
Paris-based Mistral released the weights for Mistral Large 4 on Thursday. It is a 1 trillion parameter model, it is up on Hugging Face, and the company is pitching it at businesses and governments that want to run frontier-level cyber defense on their own hardware instead of renting it from a closed lab.
The benchmark claims, all from Mistral: it narrowly beats every other open model on Artificial Analysis' Coding Agent Index, it ties China's GLM-5.3 on Terminal-Bench 4.0 at 41% (up from 28% at the October 6 preview after more reinforcement learning), and it uses 47% fewer tokens per task than GLM-5.3. For scale, Anthropic's Mythos 5.1 scored 60.9% on that same Terminal-Bench test, so the closed frontier is still well out ahead.
Pierre Stock, Mistral's VP of science, put the pitch plainly: a business or a government gets "access to frontier intelligence to defend yourselves against cyberattacks." He also admitted the obvious flip side, that "some malicious actors might try to actually leverage their own technology to attack." One knock worth knowing: during the preview, Artificial Analysis found it cost more than four times as much per task as open models of similar smarts.
What I want to know is who actually runs this thing. A trillion parameters is not something you spin up on a spare box in the closet. This is for governments and big enterprises with real GPU budgets, and that is exactly who Mistral is courting with the "sovereign" angle. I am still glad it exists. Defenders keeping a top open model in-house, instead of piping their incident data to somebody else's API, is a good trade. Just do not let anybody tell you the price of open is zero.
Sources: Fortune, TechCrunch
03 of 04
Microsoft Booked $10 Billion From OpenAI in One Quarter
Buried in Microsoft's quarterly filing with the SEC this week: $10 billion in revenue from its commercial arrangements with OpenAI, including revenue-sharing payments, for the quarter that ended September 30. Microsoft's total revenue for the quarter was $92.3 billion, so OpenAI alone is nearly 11% of the whole company's top line.
For comparison, Microsoft pulled $24.1 billion from OpenAI across all of fiscal 2026. At this new pace it is closer to $40 billion a year. OpenAI also owed Microsoft $7.2 billion in receivables at the end of September, up from $6 billion at the end of June. And Microsoft's ownership stake slipped from about 25% in June to 24% in September, without the company saying why.
Microsoft did not split out how much of the $10 billion is revenue share versus Azure compute. Context matters here: OpenAI committed to buying an extra $250 billion in Azure services last October, and an April agreement keeps the revenue share going through 2030, subject to a cap.
The way I see it, this is the clearest look yet at how tangled these two are. One customer making up a tenth of Microsoft's revenue is a great quarter and a concentration risk at the same time. And a receivable that grows by $1.2 billion in three months is the number I would watch. That is OpenAI running a tab, and tabs eventually get paid or they get renegotiated.
Sources: Microsoft 10-Q (SEC), TechCrunch
04 of 04
OpenAI's Holding Company Partner Is Buying ADT for $15.3 Billion
Thrive Holdings, the private equity spin-out of Josh Kushner's Thrive Capital, announced Thursday it is acquiring ADT for $10 per share. That is about $7.6 billion in equity and $15.3 billion in total enterprise value, per ADT. OpenAI is a partner and investor in Thrive Holdings and supplies its models, tech, and forward-deployed engineers to the companies it owns, which number more than 75.
According to the release, "ADT will form the foundation of Thrive Holdings' smart home security business." Neither side named an actual AI product yet, only that they plan to build and launch new ones. That lands the same month Apple rolled out a home security lineup and Amazon shipped new AI cameras, so the front porch is getting crowded.
Here is my read. This is the "buy a boring business, bolt AI onto it" playbook, and an established home security company that already has the customers and the monitoring is a smart place to run it. But ADT is a company where a bad AI call means a missed break-in or a cop showing up for a raccoon. I want to see what they ship before I clap. Buying the customers was the easy part.
Source: ADT investor release