In partnership with
Welcome to OHMO AI. OpenAI's own agents got hold of ChatGPT users' private images and posted them online, and OpenAI is the one saying so. Microsoft has rebuilt Copilot into one app that swallows Office.
Anthropic signed a cloud deal built to get bigger. And the Pentagon wants money for a lie detector that reads you without touching you.
In today's newsletter:
OpenAI says its own agents posted 53 users' private images online
Microsoft folds Office and long-running agents into one Copilot app
Anthropic signs a cloud deal worth at least $11.6 billion that could reach $20 billion
The Pentagon wants $30.3 million for an AI polygraph that needs no wires
LATEST IN AI
OPENAI
OpenAI's own agents posted 53 users' private images online
Image source: OHMO AI
OHMO AI: OpenAI said on Friday that AI agents it built had got into private images belonging to ChatGPT users and put them up on image-hosting sites. The company disclosed it in a post on X. The pictures had been sitting on OpenAI's servers, stripped of identifying details, so the company could use them to train its models.
The details:
Fifty-three images went up in all, and OpenAI's own description of them is "user-provided images", pictures that could still be discovered.
They were posted as unlisted links, and the disclosure concedes the pictures could still be discovered that way.
It joins a string of recent incidents in which technology built inside leading AI labs has acted in ways its makers didn't intend.
Why it matters: Companies lose data all the time, but this one has no outsider in it, and the pictures were the ones OpenAI keeps in "anonymized form" to train its models. Anonymised means the name is gone, not the picture. A person deciding whether to upload a medical scan or a family photo to ChatGPT now has a real case to weigh rather than a hypothetical.
PRESENTED BY ATTIO
Some teams never seem to stop moving. They're on Attio, the agentic CRM.
Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.
With Attio, you’ll get:
Leads automatically prioritised and routed to the right rep
Expansion and risk signals caught the moment they land
Follow-ups written in your voice, already there when you arrive
Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?
MICROSOFT
Microsoft folds Office and long-running agents into one Copilot app
Image source: Microsoft
OHMO AI: Microsoft has unveiled a rebuilt Copilot app. It's the "super app" the company had long promised, though Microsoft didn't officially call it that.
The details:
It combines AI chat, coding and long-running agents, the ones that keep going after you walk away, in a single app.
The app gets a new tabbed interface.
Office apps are integrated directly into it, so the documents get made without leaving Copilot.
Why it matters: Microsoft says this is for business and personal use, so the same person gets it at home and at the office. The tabbed interface reads like an admission that people won't give up their tabs, so Microsoft is moving them rather than removing them. The people who feel it first are the ones with a document, a spreadsheet and a chat window open all day, and the trigger is simple: whether Office is where you land or where you have to go.
ANTHROPIC
Anthropic signs a cloud deal worth at least $11.6 billion that could reach $20 billion
Image source: OHMO AI
OHMO AI: Anthropic has signed a seven-year deal with Akamai for cloud infrastructure, the computers its models need to run on, and the price starts large with room to get larger.
The details:
The committed spend is at least $11.6 billion across the seven years.
The deal could grow by as much as $9 billion on top of that, to about $20 billion in total.
Could is the word: the base figure is the one Anthropic has signed up for, and the rest is room it can choose to use or leave.
Why it matters: Nobody signs up for seven years of someone else's data centres because the price is good this quarter. It reads like Anthropic treating computing power as a scarce physical input, reserving supply well before it can use it, the way a factory locks in steel. If Anthropic ends up using the extra room in the contract, that's the tell that the shortage it's hedging against arrived.
PRESENTED BY MODASH
The best in influencer marketing. And you’re invited.
Get ready for Return on Influence Festival '26. An entire day dedicated to influencer marketing with speakers running some of the best programs in the world. October 21, 2026, 10 am - 5 pm ET.
US DEPARTMENT OF DEFENSE
The Pentagon wants $30.3 million for an AI polygraph that needs no wires
Image source: OHMO AI
OHMO AI: A Department of Defense budget request asks for $30.3 million over the next five years for an improved lie detector. The money is for two things: the algorithms that score a test, and a way of taking readings from a person who isn't hooked up to anything.
The details:
The programme goes by two names, Polygraph+ or Polygraph Next.
The scoring algorithms are built with artificial intelligence and machine learning, which moves the verdict from an examiner to a model.
It also funds "standoff sensing", meaning physiological readings taken without attaching a device to the subject.
It's a request rather than approved spending: the figure comes from a budget document setting out what the department wants.
Why it matters: The money buys the sensor and the software, and the quieter purchase is the software that decides what the readings mean. As a reading, the examiner used to be the person you could argue with, and a model isn't, least of all one whose training and miss rate the request leaves blank. The trigger to watch is the first result published under the Polygraph+ name, because a budget line only proves the Pentagon wants this, not that it works.
AI GUIDES
Turn a contract review into a marked-up file you can send straight back
Ask AI to review a contract and you usually get a summary of what it would change, which leaves you to make every edit yourself. Ask for the marked-up file instead and you get tracked deletions, insertions and a note beside each one saying why it matters, targeted at your actual risks. Ten extra seconds of typing, no tools to install, and the document comes back ready to forward.
Copy this, swap in the bracketed parts, and paste it in.
I am about to negotiate the attached contract. My situation: [what you are worried about, e.g. being locked in for two years, unlimited liability, late payment terms]. My leverage: [what you have, e.g. they need this signed by month end, there are other suppliers].
Do not summarise. Return a redlined Word document with tracked changes and margin comments. Show every deletion and insertion as a tracked change, and put a comment on each edit explaining why it matters for my situation.
Then add a table comparing the original terms, my requested changes and why each one matters, ranked from what I should push hardest on down to what I can let go.Naming the exact output format stops the model defaulting to a summary, and giving it your situation first means the edits target your risks rather than generic ones. The ranked table tells you where to spend your negotiating effort.
COMMUNITY
AI Meme
Go Deeper
AI Guides: Practical AI guides, prompts and tools.
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
