In partnership with
Welcome to OHMO AI. OpenAI is switching off Cursor. The coding startup was bought by SpaceX, and OpenAI says it can't be confident SpaceX will keep to its terms of service, pointing back at contracts Elon Musk's companies have violated. It plans to shut off access on November 12.
Elsewhere companies signed one letter saying AI cyber attacks will get far more widespread in the coming months, and Zhipu says its new open model ran on chips made at home.
In today's newsletter:
OpenAI plans to cut Cursor off from its models on November 12
Claude beat human players in 76% of a German platform's test sessions
Companies sign a single letter on AI cyber attacks
Zhipu's viral Ox Alpha model ran on a homegrown chip cluster
LATEST IN AI
OPENAI
OpenAI cuts off the coding tool SpaceX bought
OHMO AI: OpenAI says it's ending the contract that supplies Cursor, the coding startup SpaceX acquired, and plans to stop serving those models on November 12. In a statement put out on Friday night, the company said it can't be confident SpaceX will use its technology inside the terms of service, citing its experience of Elon Musk's companies violating contracts. Nothing in that reason is about Cursor's software. It's about who owns it now.
The details:
OpenAI plans to shut the access off on November 12, and the statement went out on a Friday night.
The reason OpenAI gives is trust, not price and not a technical fault.
The complaint reaches back past Cursor, to contracts OpenAI says Musk's companies have violated.
Cursor was acquired by SpaceX recently, and OpenAI has been supplying the models behind it.
Why it matters: OpenAI plans to cut off a paying customer, and it doesn't seem worried about losing the money. That reads like a seller who expects the seat to refill itself, which only happens when more people want frontier models than can get them. Worth watching is whether that holds a quarter out: if another supplier picks Cursor up and nobody notices the difference, the leverage was borrowed rather than owned.
PRESENTED BY STORECLAW
Stop Paying for 6 Tools. One AI Does It All
Most e-commerce sellers are running their store across 6 to 8 separate tools — and paying hundreds of dollars a month for the privilege. StoreClaw replaces your entire stack with one autonomous AI engine that monitors competitors, optimizes listings, automates marketing, and tracks real profit across Shopify, Amazon, and beyond.
It doesn't wait for you to ask. It runs 24/7 in the background, so you wake up to a full dashboard instead of a list of things you forgot to check.
Connect your store, and StoreClaw gets to work — no prompts, no complex setup, no six-app stack.
Free to start. No credit card required.
SCALABLE CAPITAL
Claude beat the humans in 76% of trading sessions
Image source: OHMO AI
OHMO AI: German investment platform Scalable Capital is opening up to AI assistants including ChatGPT, Claude and Grok. Investors in Europe can look at a portfolio, place trades and set up savings plans by typing a prompt. Across roughly 200 sessions Claude came out ahead of the human players 76% of the time and ChatGPT 63%, though the study also found the AI systems took on too much risk for the trade in front of them and struggled to manage that risk on the simulated trades.
The details:
Claude won 76% of roughly 200 sessions. ChatGPT won 63% of them.
The same study found the assistants took on too much risk relative to the trade in front of them.
The trading decisions in the test were simulated, so nobody's money moved on the result.
ChatGPT, Claude and Grok are all named as assistants the platform is opening to.
Why it matters: Claude won 76% of sessions and ChatGPT 63%, and that gap decides which assistant a saver points at their own portfolio. Neither number says how much a bad session lost, because the same study found both took too much risk for the trade in front of them. Worth watching is whether the risk figures turn up beside the win rates, because right now anyone choosing between the two is choosing on the half of the result that flatters them.
AI INDUSTRY
Companies sign one warning about AI attacks
Image source: OHMO AI
OHMO AI: Companies from AI, tech, finance and other industries have signed a single open letter asking for cybersecurity to be shored up. The warning inside it is not about some distant horizon. It says that in the coming months, cyber attacks that use AI will become far more widespread and sophisticated as models around the world get more capable. Firms that compete with each other put their names on the same page.
The details:
The signatories span AI, tech, finance and other industries.
The window the letter names is the coming months, not years out.
It describes the attacks ahead as far more widespread and sophisticated than what exists now.
The cause it points to is models everywhere becoming increasingly capable.
Why it matters: Every signatory agreed the danger rises with model capability, and not one of them agreed to make models less capable. That's the gap in this: the letter's own logic points at the signatories' product plans, so it reads more like a forecast than a promise. Worth watching is whether any name on the page gates a release on the risk they've now put in writing, because until one does, the coming months are being set by the same people warning about them.
PRESENTED BY APOLLO
Your next 100 customers are already in Apollo
Find, reach, and close your perfect deals — without juggling five tools or hiring more reps.
Apollo gives you everything you need to build real pipeline, fast. From inbound to outbound, first touch to close.
All in Apollo.
ZHIPU AI
Zhipu names its viral model and the shares jump
Image source: OHMO AI
OHMO AI: The model that had been going around under the code name Ox Alpha has a real name now. Zhipu AI has launched it as GLM-5.3-Flash, an open-weight model, and says the system ran on a cluster of 100,000 chips produced inside China during a stealth trial. Its shares closed more than 12 per cent higher at HK$1,160 in Hong Kong on Thursday.
The details:
Zhipu's shares closed at HK$1,160 in Hong Kong on Thursday, up more than 12 per cent.
GLM-5.3-Flash is open-weight, so the weights themselves go out rather than sitting behind an API only.
The chip count Zhipu gives for the trial cluster is 100,000, all domestically produced.
Ox Alpha was the code name the model carried before the reveal.
Why it matters: Zhipu gave the weights away, so the more than 12 per cent jump to HK$1,160 reads like investors buying the cluster rather than the code. Zhipu says the run happened entirely on 100,000 chips made inside China, and a fleet that size training a model people queued up to try is a claim that home built silicon holds up at the top end. The people who move first on that aren't the traders, they're the lab managers who spent this year budgeting their compute around what they could legally import and now have a reference customer to cite.
AI GUIDES
Get answers about your own documents without the invented parts
Paste a contract, a report, or a long email thread into a chatbot and ask what it says, and you will get an answer for every question, including the ones the document never covers. One extra line in your prompt gives the AI permission to say the answer is not there, so the gaps show up as gaps instead of as confident sentences you have to catch yourself. Takes ten seconds, works in any chatbot, nothing to install.
Copy this, swap in the bracketed parts, and paste it in.
I'm going to give you a [document type, for example a contract, report, or email thread] and ask you some questions about it.
Answer only from the material below. If you're unsure about any aspect, or if the information isn't in what I gave you, say so instead of guessing. I would rather hear "that isn't in the document" than a plausible answer you aren't sure about.
At the end, list anything I asked that you could not answer from the material.
Here is the material:
[paste your document here]
My questions:
1. [question]
2. [question]
3. [question]By default a chatbot is tuned to produce a plausible sounding answer to whatever you ask, even when the facts are not in front of it, so silence is not an option it will choose on its own. The permission line changes that, because you have told it that saying "I don't know" is a correct answer and not a failure. Telling it to answer only from the material you pasted gives it a clear boundary for what counts as known. Saying you would rather hear the gap than a guess removes any lingering pressure to be helpful at the cost of being right. The list at the end turns whatever it could not answer into something you can see, so you know exactly which questions you still need to go check yourself.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Eleni P. from Thessaloniki, Greece:
“Every Monday used to be four hours of the same job. The cafes send their bread orders however they feel like sending them, three in a spreadsheet, most in the body of an email, one guy sends a voice note every single week, and I'd retype all of it into the production sheet by hand before the ovens went on. Now I dump the lot into ChatGPT and ask it to pull the quantities into one table by shop. It still gets the voice note wrong maybe once a month so I check that one myself. My head baker worked it out before I told him. He said I'd stopped snapping at everyone before 7am.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
AI Tools: A curated list of the AI tools worth knowing.
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
