Welcome back! An AI tool built at Oxford can now read routine CT scans and spot heart failure five years before symptoms appear, picking up changes in heart tissue that no human doctor and no existing imaging test can see. It is already being put forward for NHS rollout. Meanwhile, Florida's attorney general just opened a formal investigation into OpenAI over allegations that ChatGPT was used to plan a mass shooting, a startup valued at $300M wants to be the AI agent that runs your personal life and Amazon disclosed AI revenue for the first time.
In today’s Generative AI Newsletter:
Oxford Heart AI: Can a routine scan tell you about a heart attack you haven't had yet?
Florida vs OpenAI: Did ChatGPT help plan a campus shooting? A state attorney general wants answers.
Poke: What happens when AI agents stop managing your work and start managing your life?
Amazon AI Revenue: What does $15B in AWS AI revenue and $20B in custom chips tell you about where the real money is?
Latest Developments
AI Just Learned to See a Heart Attack Coming
Researchers at the University of Oxford have built an AI tool that predicts heart failure up to five years before it develops, using routine cardiac CT scans that patients are already getting. The system reads invisible changes in the fat surrounding the heart that indicate the muscle underneath is inflamed and deteriorating. No human doctor can see these patterns. No existing imaging test can detect them.
The details:
The Data: Trained and validated across 72,000 patients at nine NHS trusts with a decade of follow-up data. Published in the Journal of the American College of Cardiology.
The Accuracy: 86% on predicting heart failure within five years. Patients in the highest risk group were 20 times more likely to develop heart failure than those in the lowest. One in four of them did.
How It Works: Fat around the heart changes texture and composition when the muscle beneath is diseased. The AI reads those textural shifts from standard CT scan data and produces a risk score with no human input required.
NHS Rollout: The Oxford team is seeking regulatory approval to add the tool to routine cardiac CT analysis in NHS radiology departments. Around 350,000 patients are referred for these scans every year in the UK. The team is also adapting the tool to work on any chest CT scan, performed for any reason.
The Bigger Application: Over one million people in the UK live with heart failure. Globally, the figure is over 60 million. The condition is consistently diagnosed too late, often only when a patient arrives at hospital with severe damage. An early warning system built into scans that are already happening could shift the entire treatment model from reaction to prevention.
This is AI doing exactly what the technology was supposed to do. Reading data at a scale and resolution that humans cannot, then giving doctors the information they need to intervene years before damage becomes irreversible. If this rolls out across the NHS, it will be one of the first genuinely prospective AI diagnostic tools deployed at national scale in any country.
Special highlight from our network
Most AI tools are just design shortcuts. Some focus on layout automation; others help with fast drafting or rigid step-by-step guides. While they save time on visuals, you’re still left doing the heavy lifting of shaping the narrative and fixing brand inconsistencies.
Prezent AI leads our 2026 rankings because it bridges the gap between raw information and strategic communication. Prezent doesn’t just "generate slides"; it builds context-aware stories from your existing documents and notes.
The shift from design to strategy:
Narrative Intelligence: Astrid AI structures your thinking into a clear, compelling flow, not just filled templates.
Contextual Clarity: Prezent AI understands your data, industry, brand, and audience, so your message always lands.
Seamless Brand Consistency: Your brand identity is built into every slide automatically. No manual formatting.
If you’re looking for a platform that acts as a strategic partner rather than just a layout tool, it’s time to try Prezent AI.
At your level, every presentation is a leadership moment.
Try the Demo — and see the difference context makes.
Florida Is Investigating Whether ChatGPT Helped Plan a Mass Shooting
Florida's attorney general James Uthmeier announced a formal investigation into OpenAI after attorneys for a victim's family alleged that ChatGPT was used to plan a mass shooting at Florida State University in April 2025. Two people were killed and five injured in the attack. Subpoenas are incoming.
The details:
The Allegation: Court documents show the accused shooter entered more than 200 prompts into ChatGPT before the attack. Attorneys for the family of victim Robert Morales claim the chatbot provided assistance on how to conduct the shooting and even advised the shooter on how to make the firearm operational moments before he opened fire.
The Probe: Uthmeier cited ChatGPT's alleged links to criminal behaviour, child safety concerns and national security risks. His office will issue subpoenas to OpenAI as part of the investigation. The Morales family plans to sue.
OpenAI's Response: The company said it builds ChatGPT to understand intent and respond safely, that it continues improving its technology and that it will cooperate with the investigation.
The Wider Context: This follows a pattern. OpenAI has faced a lawsuit alleging ChatGPT encouraged a teenager to take his own life, a Senate Judiciary Committee hearing on AI harms to children and ongoing pressure around parental controls that were only added in September 2025. The Florida probe could become a template for other states.
200 prompts. Specific operational advice. A firearm made functional with AI assistance moments before the shooting began. If those allegations hold up, this is the most consequential legal challenge any AI company has faced. OpenAI's guardrails were supposed to prevent exactly this. The investigation will determine whether those guardrails failed or whether the chatbot was deliberately circumvented, and whether it matters legally either way.
A Startup Just Raised $300M to Be the AI Agent for Your Personal Life
Most AI agents are built for work. Poke is built for everything else. The startup just raised $10M at a $300M valuation to build agents that manage your personal life: scheduling, health tracking, fitness, photo editing, travel and whatever else you need handled.
The details:
How It Works: Text the Poke agent with a task. It handles the rest. The experience is designed to feel like messaging a personal assistant rather than navigating an app.
The Market Gap: Enterprise AI agents are everywhere. Personal agents that handle the rest of your day barely exist. Poke is betting that consumer demand for this kind of help is as large as the enterprise market, just underserved.
The Valuation: $300M on a $10M raise at this stage reflects investor confidence in the category rather than current revenue. The bet is on how quickly personal agent adoption grows once the product works.
The Timing: Perplexity just plugged its Computer agent into 12,000+ bank accounts via Plaid. Claude Cowork went generally available this week. The infrastructure for agents that operate across your entire life is being built in real time. Poke is positioning to be the consumer layer on top of it.
The interesting question is whether personal agents become a single app or a feature inside everything else. Apple, Google and Meta all have the distribution to build this into their existing platforms. Poke's bet is that a purpose-built product will outperform a feature bolted onto a bigger platform. That has worked before in consumer tech. Whether it works in agents remains to be seen.
Amazon Just Revealed Its AI Revenue for the First Time
Amazon CEO Andy Jassy published his annual shareholder letter and for the first time disclosed specific AI revenue figures. AWS's AI business has crossed $15B in annualised revenue. Amazon's custom chip division (Trainium, Graviton, Nitro) has crossed $20B. Both numbers had never been made public before.
The details:
The AWS AI Figure: $15B annualised. Jassy compared it to where AWS itself stood at the same stage of its growth, saying it is 260 times ahead of that pace.
The Chip Business: Custom silicon at $20B annually. Two unnamed customers asked to buy Amazon's entire Graviton chip supply for 2026. Amazon declined to protect access for other clients.
The Spending Defence: Amazon's $200B planned AI capex has rattled investors all year. Jassy's letter was a direct rebuttal, pairing the spending with locked-in customer demand and first-time revenue disclosure to justify the outlay.
The Future: Jassy floated the idea of selling racks of custom chips to third parties, which would put Amazon in direct competition with Nvidia on the hardware supply side.
AWS has always been Amazon's most profitable division. These numbers suggest AI is about to become the most profitable part of AWS. $15B from AI services and $20B from custom chips, with two customers willing to buy the entire supply, tells you that enterprise demand for AI compute has moved well past the experimental phase. The question now is whether Amazon can sustain margins as it scales spending to $200B.
Tool of the Day: Meow
Meow is infrastructure that lets AI agents open bank accounts, issue payment cards and manage money. If you are building agents that need to transact, Meow provides the financial rails.
Try this yourself:
Visit Meow's site and explore the documentation. If you are building AI agents that need to handle payments, billing or financial operations, Meow provides the banking layer so your agent can hold funds, issue cards and move money without you building financial infrastructure from scratch.
Light Bytes
Perplexity launched a Plaid integration connecting its Computer agent to 12,000+ bank accounts, turning it into a personal finance hub that builds budgets, net worth trackers and debt payoff plans from text prompts.
Gemini added custom interactive visualisations and notebooks to paid accounts. Ask Gemini Pro to "show me" any topic and it renders charts and data displays directly in chat.
xAI is undergoing an engineering reorg with CFO Anthony Armstrong departing and SpaceX executives being installed ahead of the company's planned IPO.
Claude Cowork is now generally available to all paid plans after its research preview period.
OpenAI has built a model with cybersecurity capabilities similar to Anthropic's Mythos and plans to release it to a small set of partners, according to Axios.
