AI News Recap: January 16, 2026
AI Weekly: Claude Cowork Arrives, GPT-5.2 Cracks 30-Year Math Problem, Pentagon Deploys Grok
AI Just Solved Problems Mathematicians Couldn't Crack for 30 Years
This week marked significant expansions in AI's practical reach. Anthropic introduced Claude Cowork for file management automation, OpenAI launched a health-focused ChatGPT experience with medical record integration, and GPT-5.2 demonstrated novel mathematical reasoning by solving long-standing Erdős problems with machine-verifiable proofs.
Table of Contents
👋 Catch up on the Latest Post
🔦 In the Spotlight
🗞️ AI News
⚙️ Featured Tools
🛡️ Privacy / Cybersecurity
🧩 NeuralBuddies Word Search
👋 Catch up on the latest post …
🔦 In the Spotlight
Anthropic Launches Claude Cowork, A Version Of Its Coding AI For Regular People
Category: Tools & Platforms
🗂️ Anthropic has introduced Claude Cowork, a file-managing computer agent that can read, edit, and create files in user-designated folders to help with everyday tasks like organizing downloads or turning receipt screenshots into spreadsheets.
🌐 Claude Cowork can also navigate websites via Claude’s Chrome plugin and connect to third-party apps such as Canva using Anthropic’s Connectors framework, enabling broader workflow automation beyond local files.
💻 The feature is in research preview and initially available only to Claude Max subscribers using the macOS app, with Anthropic emphasizing explicit folder permissions and clear instructions due to the tool’s ability to take potentially destructive actions like deleting important files.
🗞️ AI News
AI Models Are Starting To Crack High-Level Math Problems
Category: AI Research & Breakthroughs
🧮 Since the release of GPT 5.2, AI tools have played a growing role in high-level mathematics, including generating novel proofs for some of Paul Erdős’s long-standing problems with assistance from formalization tools like Harmonic’s Aristotle.
📈 At least 15 Erdős problems have been moved from “open” to “solved” since Christmas, with 11 solutions explicitly crediting AI models, and Terence Tao tracking multiple cases where models made autonomous or literature-assisted progress.
📐 The field is seeing increased use of proof assistants such as Lean to formalize results, with mathematicians and computer scientists beginning to routinely incorporate AI-driven reasoning and formalization into their research workflows.
Anthropic Blocks xAI From Using Claude Models, Escalating The AI Rivalry
Category: Business & Market Trends
⛔ Anthropic has blocked Elon Musk’s xAI from accessing its Claude models, after xAI developers allegedly used them via the Cursor coding tool in ways that could support competing AI products.
📜 The move reflects Anthropic’s contractual restrictions that prohibit using its models to build rival services, alongside stepped-up monitoring and technical guardrails to detect and prevent terms-of-use violations.
🌐 The article situates this cutoff within a broader pattern of access revocations and tightening platform controls across the AI sector, highlighting how such restrictions can abruptly disrupt developer workflows and integrations.
GPT-5.2 Just Solved A 30-Year Math Problem
Category: AI Research & Breakthroughs
🧮 GPT-5.2 Pro generated an original proof of Erdős Problem 397 about central binomial coefficients, which was then formalized in Lean by the Aristotle system and accepted by Fields Medalist Terence Tao.
🧠 The model has also solved Erdős Problems 728 and 729, with Aristotle auto-correcting gaps and producing Lean-verified, machine-checkable proofs that go beyond simple pattern-matching or literature lookup.
📊 Despite these advances, Tao notes the solved problems are “lowest-hanging fruit,” as GPT-5.2 performs strongly on competition-style math but remains far weaker on open-ended research requiring deeper insight.
OpenAI Launches ChatGPT Health To Connect Medical Records, Wellness Apps
Category: Healthcare & Biotechnology
🩺 OpenAI has launched a ChatGPT Health tab that answers health-related questions and lets users upload medical records and connect wellness apps such as Apple Health and MyFitnessPal.
🔐 ChatGPT Health operates in a separate space with enhanced privacy protections, and conversations in Health are not used to train OpenAI’s foundation models.
🌍 The feature is initially rolling out to a limited group of ChatGPT users outside the European Economic Area, Switzerland, and the UK, with plans to expand to all users on web and iOS in the coming weeks.
Generative AI Could Help You Lose Weight In The New Year
Category: Healthcare & Biotechnology
🥗 MIT Sloan research found that participants in an online weight loss program who used tailored generative AI for real-time, personalized dietary feedback achieved reductions in weight and BMI, with more individuals no longer classified as obese.
📱 The program used a generative AI-powered food analysis tool that interpreted meal photos, delivered individualized recommendations, and served as a low-cost, scalable intervention to support healthier eating without medication or surgery.
👥 The study found that while personalized AI support improved weight loss outcomes, attempts to use AI to build community actually reduced engagement, indicating that human-driven communities remain important alongside AI tools.
Pentagon Is Embracing Musk’s Grok AI Chatbot As It Draws Global Outcry
Category: Military & Defense
🛰️ Defense Secretary Pete Hegseth announced that Elon Musk’s Grok chatbot will join Google’s generative AI engine inside Pentagon networks to process vast amounts of military and intelligence data across classified and unclassified systems.
🌐 The deployment comes amid global backlash over Grok’s role in generating highly sexualized deepfake images, which has led Malaysia and Indonesia to block the chatbot and triggered a UK safety regulator investigation.
⚖️ Hegseth signaled a more aggressive AI posture than the prior administration, saying Pentagon AI should operate “without ideological constraints” while it remains unclear whether earlier limits on certain national security uses of AI are still in force.
DeepSeek To Launch New AI Model Focused On Coding In February, The Information Reports
Category: Tools & Platforms
💻 Chinese startup DeepSeek plans to launch its next-generation V4 AI model in mid-February, designed with strong coding capabilities.
🧪 Internal tests indicate DeepSeek V4 could outperform rivals such as Anthropic’s Claude and OpenAI’s GPT series on coding tasks.
📏 V4 is reported to handle extremely long coding prompts more effectively, which could benefit developers working on complex software projects.
Generative AI Tool Helps 3D Print Personal Items That Sustain Daily Use
Category: Generative AI & Creativity
🧩 MIT CSAIL’s MechStyle system lets users upload or select 3D models of everyday items, then use text or image prompts to stylistically personalize them while preserving their intended function.
🏗️ MechStyle couples generative AI with finite element analysis and an adaptive scheduling strategy to simulate stresses, selectively re-run physics checks, and ensure stylized designs remain structurally viable for real-world use.
🧴 The tool successfully produced durable, personalized objects such as wall hooks, pillboxes, lampshades, and assistive devices like finger splints and utensil grips, with tests showing up to 100 percent of generated designs staying mechanically sound.
Doctors Think AI Has A Place In Healthcare — But Maybe Not As A Chatbot
Category: Healthcare & Biotechnology
🩺 Clinicians acknowledge that patients already rely heavily on tools like ChatGPT for medical advice, and OpenAI’s new ChatGPT Health aims to formalize this use with a more private, safeguarded experience for health queries and record uploads.
🧾 Many doctors see greater immediate value in AI systems that reduce administrative burdens—such as Stanford’s ChatEHR and Anthropic’s Claude for Healthcare—which help clinicians navigate electronic health records and automate tasks like prior authorization requests.
⚖️ Experts highlight tensions between improved access to care via AI and concerns over data privacy, hallucinated medical advice, and differing incentives between profit-driven tech companies and patient-focused healthcare providers.
McKinsey Tests AI Chatbot In Early Stages Of Graduate Recruitment
Category: Workforce & Skills
🤖 McKinsey is piloting an AI chatbot for the initial stages of graduate recruitment, asking applicants to interact with it as part of their early assessment while keeping interviews and final decisions with human recruiters.
📊 The chatbot is used to handle high application volumes by asking consistent questions, collecting structured responses, and giving hiring teams more information to review before advancing candidates.
⚖️ The firm acknowledges concerns about fairness and bias, emphasizing that the tool is monitored, used alongside human review, and subject to checks on how it evaluates responses and impacts different candidate groups.
🔗 Read Full Article
⚙️ Featured Tools
Introducing Anthropic Labs
Category: Tools & Platforms
🧪 Anthropic is expanding Labs, a team dedicated to incubating experimental products at the frontier of Claude’s capabilities, turning early prototypes into scalable offerings.
🧱 Labs builds on recent launches such as Claude Code, the Model Context Protocol (MCP), Skills, Claude in Chrome, and Cowork, which moved from research previews to widely used products and standards.
🧑💻 Mike Krieger is joining Labs to build alongside Ben Mann, while Ami Vora becomes head of Product, partnering with CTO Rahul Patil to scale Claude experiences for millions of users.
Personal Intelligence: Connecting Gemini To Google Apps
Category: Tools & Platforms
🔗 Google is launching Personal Intelligence, a beta feature in the U.S. that lets users connect Gemini to apps like Gmail, Google Photos, YouTube, and Search to provide more personalized, context-aware assistance.
🔐 The system keeps app connections off by default, allows granular control over which apps are linked, and does not train directly on users’ Gmail or Photos libraries, instead using connected data only to answer specific requests.
📲 Personal Intelligence is rolling out over the next week to eligible Google AI Pro and AI Ultra subscribers in the U.S., working across web, Android, and iOS, with plans to expand to more countries and tiers over time.
New Tech And Tools For Retailers To Succeed In An Agentic Shopping Era
Category: Tools & Platforms
🧩 Google is launching the Universal Commerce Protocol (UCP), an open standard that lets AI shopping agents, retailers, and payment providers interoperate across discovery, checkout, and post-purchase, starting with AI-powered checkout on eligible Google product listings in Search and the Gemini app.
🛍️ A new Business Agent feature will let eligible U.S. retailers deploy branded AI agents directly in Search to answer product questions, surface recommendations, and eventually support agentic checkout and personalized offers using their own data.
💸 Google is piloting Direct Offers in AI Mode, enabling advertisers to surface targeted deals like discounts or free shipping alongside AI-powered product results, while expanding Merchant Center attributes to make retailer catalogs more discoverable in conversational shopping experiences.
🔗 Read Full Article
🛡️ Privacy / Cybersecurity
Fintech Firm Betterment Confirms Data Breach After Hackers Send Fake Crypto Scam Notification To Users
Category: AI Safety & Cybersecurity
🕵️ Hackers accessed some Betterment systems via a social engineering attack on third-party marketing and operations platforms, exposing customer names, contact details, and dates of birth.
⚠️ Using this access, attackers sent a fraudulent in-app notification promising to triple users’ crypto if they transferred $10,000 to a wallet controlled by the hackers.
🛡️ Betterment says it revoked unauthorized access the same day, is investigating with a cybersecurity firm, and reports that no customer accounts, passwords, or login credentials were compromised, though it has not disclosed how many customers were affected.
Phishing Campaign Abuses Google Cloud Services To Steal Microsoft 365 Logins
Category: AI Safety & Cybersecurity
📧 Attackers are sending realistic phishing emails from a legitimate-looking Google address using Google Cloud Application Integration’s Send Email feature, luring victims with routine prompts like voicemail or document notifications.
🔗 The emails link to genuine Google Cloud Storage and googleusercontent domains, then redirect users to a fake Microsoft 365 login page where any entered credentials are harvested by the attackers.
🛡️ Google has blocked several related campaigns and stressed that the activity exploits its workflow automation tools rather than any breach of its infrastructure, while advising ongoing user vigilance against spoofed brand emails.
Executives Worry Most About Cyber-Enabled Fraud, Geopolitics And AI
Category: AI Safety & Cybersecurity
💸 A World Economic Forum survey finds cyber-enabled fraud is now executives’ top cyber risk, ahead of ransomware, with 73% saying such fraud affected them or someone they know in 2025.
🤖 AI-related vulnerabilities have entered CEOs’ top concerns alongside software flaws and supply-chain threats, with particular worries about data leaks, hackers’ advances, and security weaknesses in AI systems.
🌍 Two-thirds of CEOs say geopolitical volatility has changed their cyber strategies, and nearly 60% of respondents view cyber regulations as helpful for raising security awareness and improving defenses, despite added compliance complexity.
🔗 Read Full Article
🧩 NeuralBuddies Word Search
