In today’s edition of Data Points, you’ll learn more about:
- Anthropic’s play for the healthcare market
- UCP, a new standard for agentic shopping
- Malaysia and Indonesia’s stand against Grok
- DeepSeek’s new memory management method
But first:
Cowork is Claude Code for everything besides code
Anthropic released Cowork, a new interface that gives Claude access to local folders where it can read, edit, and create files to complete multi-step tasks. Available today as a research preview for Claude Max subscribers on macOS, Cowork builds on the same foundations as Claude Code but targets non-coding workflows like organizing downloads, extracting data from screenshots into spreadsheets, or drafting reports from scattered notes. The system works with existing Claude connectors and new document creation skills, and users can queue multiple tasks for Claude to handle in parallel. Anthropic warns that Claude can take potentially destructive actions like deleting files if instructed, and the system remains vulnerable to prompt injection attacks despite built-in defenses. (Anthropic)
Apple taps Google’s Gemini to power Siri upgrades and AI features
Apple announced a multiyear partnership with Google to use Gemini models and cloud infrastructure for Apple Foundation Models, with an upgraded Siri expected in early 2026. The models will run on Apple devices and Apple’s private cloud compute. The companies did not disclose financial terms, but Bloomberg previously reported Apple planned to pay roughly 1 billion dollars annually for Google AI access. Apple delayed its planned Siri AI upgrade from 2025 to 2026 and currently uses OpenAI’s ChatGPT for complex queries through Siri and Apple Intelligence, an arrangement the company says will continue unchanged. Nevertheless, Apple choosing Google shows the latter companies’ gains in broad AI capabilities, and Apple’s relative inability to continue to competitively develop its own AI products. (CNBC)
Claude for Healthcare features medical skills and connectors
Anthropic’s Claude for Healthcare is a HIPAA-compliant version of its AI assistant that connects directly to medical databases including the CMS Coverage Database, ICD-10 codes, and the National Provider Identifier Registry. The new service aims to speed up prior authorization reviews, claims appeals, and care coordination tasks that currently consume significant administrative time. Individual subscribers in the U.S. can also connect their health records from HealthEx, Function, Apple Health, and Android Health Connect to get plain-language explanations of test results and medical history. Users control exactly what data they share and Anthropic commits not to use health data for model training. Healthcare administrative costs in the U.S. exceed 250 billion dollars annually, and tools that streamline these processes could materially reduce costs and improve patient access to care. (Anthropic)
Google introduces universal standard for AI-powered shopping
Google launched the Universal Commerce Protocol, an open standard that enables AI agents to complete shopping tasks across discovery, purchase, and support. The protocol works with existing standards like Agent2Agent and Agent Payments Protocol and was developed with Shopify, Etsy, Wayfair, Target, and Walmart, with endorsements from more than 20 companies including American Express, Mastercard, Stripe, and Visa. Google will use UCP to power checkout directly within AI Mode in Search and the Gemini app for eligible U.S. retailers, with Google Pay and PayPal as payment options. The company also introduced Business Agent, which lets retailers deploy branded AI assistants on Search to answer product questions, and Direct Offers, a pilot program that allows advertisers to present exclusive discounts within AI Mode search results. (Google)
Asian nations block Grok over explicit AI images
Malaysia and Indonesia became the first countries to block Grok, the AI chatbot developed by Elon Musk’s xAI, citing widespread misuse to generate non-consensual sexual images and deepfakes. Both nations’ regulators found that existing safeguards failed to prevent the creation and distribution of fake pornographic content involving women, minors, and real people without their consent. Indonesia’s Communication and Digital Affairs Minister described non-consensual sexual deepfakes as a serious violation of human rights and citizen safety. The temporary blocks will remain in place until xAI implements effective safeguards, though the company has so far responded only with user reporting mechanisms. Grok, which launched in 2023 as a free tool on X and added an image generator with “spicy mode” for adult content last summer, now faces mounting international scrutiny from the EU, Britain, India, and France. (Associated Press)
DeepSeek Engram paper details method to improve knowledge recall
Researchers from DeepSeek and Peking University introduced Engram, a memory lookup system that stores static knowledge in hash tables rather than forcing language models to reconstruct it through computation. The module uses N-gram embeddings retrieved through constant-time lookups, scaling to 27 billion parameters while maintaining computational efficiency equivalent to smaller models. The system offloads its massive embedding tables to host memory during inference, using deterministic addressing to prefetch data with less than three percent overhead. Researchers found that by handling routine pattern matching through lookups, Engram frees early neural network layers for complex reasoning and allows attention mechanisms to focus on long-range context rather than local dependencies. When tested against models with identical parameters and compute requirements, Engram delivered improvements of five points on reasoning benchmarks, 3.4 points on knowledge tests, and three points on coding tasks. (arXiv)
Still want to know more about what matters in AI right now?
Read the latest issue of The Batch for in-depth analysis of news and research.
Last week, Andrew Ng discussed a new course designed to teach non-coders how to build AI-driven applications, emphasized the importance of coding skills for productivity, and encouraged everyone to engage with AI tools.
“If you take this course, you will build a working web application: a funny interactive birthday message generator that runs in your browser and can be shared with friends. You’ll customize it by telling AI how you want it changed, and tweak it until it works the way you want.”
Read Andrew’s letter here.
Other top AI news and research stories covered in depth:
- OpenAI fine-tuned a version of GPT-5 to confess when it was breaking the rules, aiming to improve model transparency and trust.
- SAIL introduced the Science Context Protocol, a lingua franca for AI agents to communicate effectively about local and virtual experiments.
- A Microsoft study revealed that Copilot’s users changed hour to hour, highlighting the dynamic nature of AI interaction across different times and devices.
- Canadian researchers discovered that capping context helped models better retrieve data, making AI reasoning more affordable and efficient.
A special offer for our community
DeepLearning.AI recently launched the first-ever subscription plan for our entire course catalog! As a Pro Member, you’ll immediately enjoy access to:
- Over 150 AI courses and specializations from Andrew Ng and industry experts
- Labs and quizzes to test your knowledge
- Projects to share with employers
- Certificates to testify to your new skills
- A community to help you advance at the speed of AI
Enroll now to lock in a year of full access for $25 per month paid upfront, or opt for month-to-month payments at just $30 per month. Both payment options begin with a one week free trial. Explore Pro’s benefits and start building today!