OpenAI's Math Scandal, AI Power Outages & Clearview's Dossier Tool
September 10, 2026The AI industry is fighting about math, electricity, surveillance, and whether anyone is actually in charge of any of it.
OpenAI Has a Math Problem — and It Might Be an Ethics Problem
A second mathematician has gone public accusing OpenAI of training its models on unpublished research without permission. This comes just days after a separate, very public dispute over the same issue. For context: AI models learn by ingesting enormous amounts of text, and unpublished academic work — papers shared privately between researchers — would be unusually valuable training material because it contains cutting-edge reasoning that hasn't circulated widely yet. If OpenAI used that work without consent, it would be both an intellectual property issue and a strategic own-goal: researchers control access to the frontier work AI labs need most, and academics who feel burned don't share preprints, don't collaborate, and do talk to regulators. OpenAI hasn't confirmed or denied the specifics, which means every new accusation lands in a vacuum — and right now, a second mathematician going public in the same week is starting to look less like a dispute and more like a pattern.
The Dirty Secret of AI Is That It Keeps Breaking the Power Grid
On July 22nd, a single transmission line fault in Ashburn, Virginia — home to the densest concentration of data centers on the planet — yanked more than 3 gigawatts of electricity off the grid in seconds. To put that in perspective, 3 gigawatts is roughly what it takes to power 2.5 million American homes. This wasn't a freak event; it's the second major incident in two years from the same region. The core issue is that AI data centers are enormous, sudden consumers of power, and the electrical infrastructure wasn't built to absorb that kind of demand in that kind of concentration. If your business depends on cloud services — and at this point, most do — this is the failure mode that hits you before you see it coming: the data center goes dark, your cloud provider goes with it, and your operations stop while your IT team scrambles to explain something nobody was tracking.
https://www.technologyreview.com/2026/09/10/1141649/powering-ai-is-an-architecture-problem/
Clearview Just Got a Lot More Dangerous
Clearview AI — the company that built a face recognition tool allowing law enforcement to match photos against billions of scraped images — is now testing something called InquiryIQ, a prototype that goes several steps further. Feed it a face, and it uses an AI model (in this case from Elon Musk's xAI) to surface that person's social media accounts, known associates, and other online information automatically. Previously, Clearview identified who you were. This new tool attempts to map out your entire online life from a single photo. Think about what that means for a journalist whose source attended a public conference, or a therapist whose client shows up in a tagged photo — one image becomes a dossier. The legal frameworks governing this kind of tool are far behind the technology.
OpenAI Adds the Guy Who Worries AI Will Kill Us All to Its Board
Paul Christiano, one of the most respected researchers in AI alignment — the field concerned with ensuring AI systems actually do what humans intend, rather than finding dangerous shortcuts — is joining the OpenAI Foundation's board of directors. Christiano has publicly estimated a meaningful probability that advanced AI leads to catastrophic outcomes for humanity. OpenAI appointing him is interesting for two reasons: it signals that the "we should be genuinely scared" faction now has a formal seat at the table, and it creates useful cover for the company to say it takes safety seriously at a moment when it's facing scrutiny from every direction. Whether it changes anything internally is a different question.
https://techcrunch.com/2026/09/09/openai-adds-a-prominent-ai-doomer-to-its-board-of-directors/
A $1.5 Billion Round, Abandoned
Listen Labs, an AI research startup focused on qualitative research tools, walked away from a signed $1.5 billion Series C term sheet — a formal agreement to receive funding — in order to pursue acquisition talks with Salesforce instead. Walking away from a signed term sheet is unusual and carries real legal and reputational risk with investors. The fact that they did it anyway suggests either the Salesforce offer is substantially better, or the founders see the acqui-hire path as more certain than trying to build independently in an environment where large platforms are eating the market from underneath startups. For anyone in sales, marketing, or operations who uses or evaluates research tools, this is worth watching — products that get acquired tend to get absorbed, repriced, or discontinued.
The pattern across today's news is that AI's biggest problems right now aren't technical — they're about who owns the inputs, who controls the infrastructure, who has access to the surveillance, and who, if anyone, is accountable when something goes wrong.
Stop staring at the blank ChatGPT box
The Plain-English AI Prompt Pack: 100 copy-paste prompts for real work — emails, meetings, marketing, managing people. Works with any chatbot.
Get the Prompt Pack →Get the daily brief
Plain-English AI news for people with real jobs. Free, three minutes a day.