In partnership with
Welcome to OHMO AI. OpenAI has stopped taking new customers for its priciest ChatGPT plan, and it isn't for lack of demand. Apple's rebuilt Siri is launching, and which phone you hold decides what you get.
Anthropic says Chinese labs trained on Claude, and it has put a number on it. Oracle's AI cloud is growing fast, on borrowed money.
In today's newsletter:
OpenAI stops taking new $200 ChatGPT Pro subscribers
Apple's rebuilt Siri is coming as a beta later this year, English only and not for older iPhones
Anthropic says Chinese AI labs harvested Claude across nearly 200 million exchanges
Oracle grows 30% on AI cloud demand while its debt hits $125 billion
LATEST IN AI
OPENAI
OpenAI stops taking new $200 ChatGPT Pro subscribers
Image source: OHMO AI
OHMO AI: OpenAI has stopped taking new sign-ups and upgrades for ChatGPT Pro, the plan that costs $200 a month and is where Astra is available, because there isn't enough capacity to serve everyone who wants it. Head of product Thibault Sottiaux wrote on X that this plan puts the most strain on the company's systems.
The details:
Pro comes with 20 times the usage of the Plus tier below it, and it's the most expensive plan OpenAI offers.
The company announced the halt on Thursday and calls it temporary.
Sottiaux, who leads product for Codex and ChatGPT, framed the pause as the option that disrupted the least.
He wrote that the company wanted to take the smallest step that would let it keep giving the broadest access possible.
Why it matters: Refusing money instead of raising the price reads like a company that has run out of machines rather than customers. Sottiaux's "broadest access" line says what the trade was: fewer heavy users so more light ones keep running. Anyone who wanted to pay for Astra this week now waits, and how long the pause lasts is the plainest measure of how short the compute is.
PRESENTED BY ROKU
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.
LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.
The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.
Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.
The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.
APPLE
Apple's rebuilt Siri is coming as a beta that older iPhones won't get
Image source: OHMO AI
OHMO AI: Apple's Siri AI, the rebuilt assistant developers have been testing, is coming to users as a beta later this year. It runs on Apple Intelligence and will pull answers from the web and from your own messages, emails and photos, and it will know what's on your screen. It'll be English only at first, and it will need at least an iPhone 15 Pro or an iPhone 16.
The details:
Only developers have it today, and it will reach users as a beta later this year.
The most advanced model will need an iPhone 17 Pro or newer, so older phones get a lesser version of the assistant.
It will include a dedicated app for revisiting conversations across your Apple products, plus an expanded Visual Intelligence experience and integrated writing tools.
If you're in the European Union you'll wait longer, while Apple addresses the EU's privacy and security concerns.
Why it matters: None of this is on a user's phone yet. Developers are testing it now, and the beta people will get later this year is English only at first. What changes today is the upgrade maths. A person on anything older than an iPhone 15 Pro will get no Siri AI at all, a person on an iPhone 15 Pro or an iPhone 16 will get it without its most advanced model, and only an iPhone 17 Pro or newer will get the full version. Each of those people now reads the same announcement as a different price.
ANTHROPIC
Anthropic says Chinese AI labs harvested Claude across nearly 200 million exchanges
Image source: OHMO AI
OHMO AI: Anthropic has published a report accusing China-based AI labs, Alibaba, Moonshot and DeepSeek among them, of running large-scale distillation attacks on Claude, meaning they fed its answers into their own models as training data. It counts nearly 200 million exchanges across five separate campaigns, says it detected and disrupted them, and says the methods have grown more sophisticated in recent months as competition has intensified.
The details:
Anthropic says operators affiliated with Alibaba used Claude outputs to help train its Qwen models. It calls the Alibaba operation the largest distillation campaign it has measured, involving more than 151 million exchanges with Claude between May and July.
Moonshot took a different route: it routed some requests from its own Kimi users to Claude and trained on some of the replies. In one 10-day period it relayed nearly 300,000 customer requests, the vast majority of them going to Claude Opus models.
The report says the campaigns went after Claude's most valuable capabilities, naming "agentic capabilities and tool use, coding and data analysis, and logical reasoning".
Anthropic first spoke out about distillation attacks in February and named specific labs then. OpenAI has reported the same kind of activity and pinned it on DeepSeek.
Why it matters: Every figure here is Anthropic's count, from Anthropic's logs, on a day Anthropic picked, and the campaign it calls the largest ran from May to July while the report landed on Thursday. That reads like an account of what one company caught, with nothing in it that lets an outsider check what it missed or why it waited. The tell will be whether Alibaba, Moonshot or DeepSeek answer it in public, and with what.
PRESENTED BY ARTISAN
Hire Ava, the AI BDR built for enterprise
Ava is the first AI BDR to run outbound end to end, finding leads or ingesting your CRM accounts, sending personalized emails on your reps' behalf, and booking meetings, autonomously or on copilot. She runs outbound for DoorDash and Grammarly. She's SOC 2 Type II audited, SSO and GDPR ready.
ORACLE
Oracle grows 30% on AI cloud demand while its debt hits $125 billion
Image source: OHMO AI
OHMO AI: Oracle has reported 30% revenue growth for its latest quarter, topping estimates, with the lift coming from renting out cloud infrastructure for AI workloads. Cloud revenue rose 62% from a year earlier to $11.6 billion. The same results show $125 billion in debt as the company pays for a sweeping buildout of data centres, and its shares whipsawed on Friday.
The details:
Total revenue reached $19.35 billion, up from $14.93 billion a year earlier and ahead of the $19.14 billion analysts expected.
Cloud infrastructure, the business that rents AI servers to enterprise customers, grew 121% to $7.4 billion and beat Wall Street's expectations.
Cloud applications grew at a slower pace than infrastructure, and Oracle's software revenue line slipped.
Free cash flow came in at negative $5.4 billion for the quarter, against negative $362 million a year earlier.
Why it matters: The growth is the headline, but the debt is the story: $125 billion carried to build data centres whose rental income has to outrun the interest on it, while free cash flow went far deeper into the red than a year earlier. That reads like AI servers bought with borrowed money now and paid for by tenants later. If enterprise customers keep signing to rent them, the beat holds. If they start haggling, Oracle is the one left holding the building.
AI GUIDES
Get a review you can trust by setting the rules before you paste the work
Paste a draft and ask if it looks OK and AI grades it against whatever it feels like that day. Tell it which categories to check, which rules are non-negotiable and the exact fields you want back, and it grades against your standard, one finding at a time, so you get a list you can act on instead of an opinion you have to argue with. Two minutes to set up, works in any chatbot.
Copy this, swap in the bracketed parts, and paste it in.
Review the material at the end of this message against the standard below. Read the whole brief before you start.
The standard: [paste or describe the rules, for example your brand voice notes, a style guide, the client's requirements or your house rules]
Check only these categories: [list three to five, for example tone, claims that overpromise, missing required lines, wrong names or dates]
Hard rules, always a problem if broken: [for example the legal line must appear, prices must match the price list]
Acceptable variation, do not flag: [for example sentence length, order of sections, casual greetings]
For every problem you find, give me exactly these fields:
- Where: the sentence or section
- Issue: what is wrong, in one line
- Should be: what the standard says it should match
- Confidence: high, medium or low
List hard rule breaks first. Do not mention anything that is fine. If the standard does not cover something you noticed, say so in one line at the end instead of flagging it.
Material to review:
[paste it]Naming the categories and the hard rules gives the model a fixed standard to weigh each finding against, instead of guessing at what counts as wrong. Fixing the fields per finding turns a loose opinion into a list you can check and act on.
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 Mattias L. from Karlstad, Sweden:
“Boiler room, hands still wet, heat pump showing a code I've seen twice in ten years. Photo of the display and the plate into chatGPT on my phone. Then, fika in the van, I type this: Here's the error code and the model plate. Tell me what this code means on this unit and the three things that usually fail around it, most common first. Then write the note for the customer: what was wrong, what I did, why the bill is what it is. Plain Swedish, five lines, no jargon.
Back comes a parts list I'd have got to anyway, twenty minutes sooner. Newer models it doesn't know, so I paste the manual page first. The note goes on the invoice before they Swish me. Too polite every time. I cut the sorry from the first line myself.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
