Today's AI Pulse
Anthropic shuts down Claude Fable 5 and Mythos 5 after US export‑control warning
Anthropic disabled access to its Claude Fable 5 and Claude Mythos 5 models after the Trump administration, via the Commerce Department, warned that foreign nationals could exploit the systems. The move followed alerts from Amazon and other firms about a potential jailbreak that could aid cyber‑security attacks. Anthropic declined the administration’s request for a voluntary shutdown but complied with the order.
Also developing:
By the numbers: AMD acquires AI memory‑optimization pioneer Mext
OpenAI Knew. It Chose Not to Call the Police. Now Sam Altman Is Sorry.
Sam Altman publicly apologized to the Tumbler Ridge community after OpenAI failed to notify police about a ChatGPT user who was flagged in June 2025 for discussing gun‑violence scenarios. Internal reviews recommended reporting the threat, but leadership applied a higher threshold and did not contact law enforcement. The shooter later killed eight people and injured 27 in Canada’s deadliest school shooting in decades. OpenAI has since lowered its reporting threshold, established a direct RCMP liaison, and announced voluntary policy changes, though Canada has no legal requirement for AI firms to report such threats.
AI Helps Chemists Design Molecules Step by Step
Researchers at EPFL unveiled Synthegy, a new framework that pairs large language models with traditional retrosynthesis and mechanism‑prediction tools. By translating candidate pathways into text, the LLM evaluates each route against plain‑language user goals and scores its chemical plausibility. In...
The AI Rally Looks Like the Dot-Com Bubble. The Companies Do Not.
The S&P 500’s Shiller CAPE ratio has climbed to 38‑40, the second‑highest level in 155 years and only behind the dot‑com peak of 44.19. At the same time, the ten largest stocks now own roughly 38% of the index, a concentration about...
Google Cloud Uses Travel to Show What Agentic AI Can Actually Do
At Google Cloud Next, the company unveiled its Gemini Enterprise Agent Platform, a mission‑control layer for coordinating multiple AI agents across data and workflows. The platform was demonstrated using the travel sector, with Virgin Voyages launching Rovey, an AI assistant that...
From Rule of Thumb to Mechanistic Formula: An AI-Assisted Model for Individualizing Shuttle Run Distance in HIIT
A new AI‑assisted mechanistic formula replaces the long‑standing 0.7‑second rule for adjusting shuttle‑run distance in high‑intensity intermittent training (HIIT). The model computes time lost per change of direction (COD) as t_cod = v / A0, where A0 reflects each athlete’s acceleration and deceleration capacity,...
AI Twins Boost HR Efficiency, Threaten Talent Acquisition Jobs
The AI Twin Is Learning Your Job SHRM's 2026 AI in HR full report finds 89% of AI-adopting organizations report operational efficiency gains, and 27% have deployed AI deepest in talent acquisition: the function AI can replace most cleanly. The productivity...
The Future Is Agentic in Recommender Systems
In this episode of Data Skeptic, associate professor Yashar Delju discusses how large language models (LLMs) are reshaping recommender systems, moving from classic collaborative‑filtering pipelines to agentic, conversational approaches. He explains the emerging trustworthiness challenges—adversarial attacks, fairness, robustness, privacy, and...
Participatory Modelling and Simulation to Improve AI-Based Public Social Services
The new open‑access volume presents a suite of agent‑based simulations that examine AI’s role in welfare‑related public services across nine countries. It emphasizes participatory modelling to ensure AI systems are context‑specific, adaptive, and aligned with diverse societal values. Case studies...
Experian Takes Consumer Financial Guidance to the Next Level with AI-Powered Virtual Assistant
Experian has launched version 3.0 of its Experian Virtual Assistant (EVA), an AI‑powered chatbot that moves beyond credit‑score explanations to analyze spending, flag recurring subscriptions and even execute actions such as credit freezes. The assistant relies on generative AI models but...
10 Prompts for the New Gemini Agent
Google’s Gemini Agent extends beyond text generation by executing multi‑step tasks across Workspace apps with live web browsing. The blog shares ten customizable prompts that let the agent handle procurement research, email drafting, and inbox triage, all while pausing for...
AI Demand Is Still Booming
AI spending and token consumption are accelerating at unprecedented rates. SemiAnalysis lifted its AI budget to a $7 million annualized run‑rate, while Anthropic’s annual recurring revenue surged to $35‑45 billion with margins above 72%. The rapid growth is outpacing compute supply, as...
U.S. News Sues OpenAI, Adding to the Publisher AI Copyright Wave
U.S. News & World Report has filed a lawsuit against OpenAI in the Southern District of New York, alleging the AI developer used the publisher’s articles without permission to train its models and generate competing outputs. The case joins a...
AI Reliability Engineer: The Future Code Orchestrator
There is a new role is that is needed in the world of AI: The AI Reliability Engineer. The developer of the future will be an orchestrator of agents, a master at understanding code management, a new breed that can...
How Anthropic’s Mythos Model Is Forcing the Crypto Industry to Rethink Everything About Security
Anthropic’s Mythos AI model is prompting DeFi firms to broaden security beyond smart‑contract code to the underlying infrastructure such as key‑management, bridges and oracle networks. By simulating adversarial behavior and chaining minor flaws, Mythos can expose systemic, cascading failures across...
Remodex Is the Best Codex Remote Client for iOS (Until OpenAI Releases an Official Codex Mobile App)
Remodex is an indie iOS client that lets developers control the OpenAI Codex CLI running on a remote Mac. The app connects via a lightweight bridge and Tailscale, using a QR code to pair the iPhone with the Mac studio....
Enterprises Question Trust in Anthropic After Recent Issues
I attended Google Cloud Next this week and @AnthropicAI had a giant spread there. Booths, of course, but also rented out a swanky restaurant. After what they’ve pulled recently with performance, price and quality half-truths, I was thinking about enterprise...
Human and AI Intelligence Occupy Orthogonal Planes, Not a Line
🚨Intelligence isn’t a line. It’s a plane. Human cognition and AI computation don’t share the same axis--they run on orthogonal vectors. 👉 This is the heart of anti-intelligence. AI delivers scale, speed, pattern fluency and statistical reach… but has no embodiment,...
Salesforce Unveils Headless 360, a No‑Login AI‑Agent Platform for Enterprise Sales
Salesforce introduced Headless 360, a platform that exposes its entire CRM stack as APIs, MCP tools and CLI commands, allowing AI agents to operate without a browser login. The overhaul, built over two and a half years, adds more than...
Apple's Incoming CEO John Ternus Faces Pressure to Launch Breakthrough AI Product
Apple’s soon‑to‑be CEO John Ternus is under intense scrutiny to roll out a flagship AI device that can sustain the company’s brand momentum after Tim Cook’s September exit. Industry analysts warn that without a breakthrough, Apple risks losing its edge...
COAI DigiCom Summit Calls for AI‑Led Networks and Tougher Anti‑Fraud Measures in India
At the COAI DigiCom Summit 2026, regulators and industry leaders pressed for AI‑led telecom networks, a hybrid fiber‑and‑FWA broadband model and tighter anti‑fraud systems. The push comes as fixed‑line subscriptions reach 63 million (about 20% of households) and India eyes a...
Equifax Warns AI‑driven Fraud Could Hit $40 B by 2027, $15.9 B Lost in 2025
Equifax warned that AI‑enabled fraud cost U.S. consumers a record $15.9 billion in 2025 and could climb to $40 billion by 2027. The agency highlighted synthetic identities, deepfakes and cheap AI toolkits as the new weapons reshaping financial crime.
Keep Experience Adaptive, Not Fixed, Thanks to AI
This gets at something I see all the time. Experience only stays valuable when it keeps updating. AI lowers the cost of trying, learning, and changing your mind. That makes stale judgment easier to spot. The most effective operators treat experience like a...
SUSE Launches Model Context Protocol Ecosystem to Power AI‑Driven Infrastructure Ops
SUSE announced that its Model Context Protocol (MCP) is now generally available across Rancher Prime, Multi‑Linux Manager, SUSE Linux and SUSE AI, backed by a partner ecosystem that includes AWS, Fsas Technologies, n8n, Revenium and Stacklok. The move gives enterprises...
Norm AI Unveils Legal AGI Lab to Craft Compliance Rules for Autonomous AI
Norm AI announced the creation of a Legal AGI Lab dedicated to building legal and compliance structures for autonomous AI systems. CEO John Nay said the lab will tackle accountability, liability and governance as AI agents move into economically critical...
AI Safety Can Be a Pascal's Mugging Even if P(doom) Is High
The article argues that labeling AI safety as a Pascal’s mugging is misguided because the concept depends on the probability an individual can make a difference, not on the baseline risk of catastrophe. Even if the chance of AI doom...
Sony AI’s Project Ace Robot Beats Elite Table‑Tennis Players, Showcasing Next‑Gen Robotics Hardware
Sony AI’s Project Ace robot defeated three of five elite table‑tennis opponents in a sanctioned match, proving that a combination of high‑speed perception, reinforcement‑learning control and custom hardware can outplay human experts. The win signals a turning point for embodied...
YouTube Launches Free Deep‑Fake Detection Tool for Hollywood Celebrities
YouTube is extending its free likeness‑protection service to Hollywood actors and musicians, letting them scan for AI‑generated deepfakes and request takedowns. The move follows the platform’s earlier rollout to journalists and political candidates and comes amid rising concerns over synthetic...
What Shopify Merchants Need to Know About AI Generated Fakes in 2026
Shopify merchants generating $500 K‑$20 M in sales are now facing a surge of AI‑generated fraud, with Pindrop reporting a 1,210% rise in AI‑driven attacks by December 2025. The threat spans five distinct forgery types—fake reviews, fabricated damage photos, synthetic testimonials, brand impersonation,...
3 Prompts, 1 Project, and 1 Claude Skill to Help You Actually Execute What You Create with AI
The post argues that AI users, especially Claude fans, hoard prompts and half‑finished projects without delivering results. It introduces a three‑layer framework—Jobs‑To‑Be‑Done, Theory of Constraints, and CBT Reframing—plus a Claude “Project” skill that audits what you’ve actually shipped. By turning...
Using Coding Assistance Tools to Revive Projects You Never Were Going to Finish
The author used Claude Code, an AI coding assistant, to rebuild a personal FastAPI shim that exposes YouTube Music through the OpenSubsonic API. Starting from a minimal repository, Claude generated stubbed endpoints, added search via ytmusicapi and streaming via yt‑dlp,...
Norway Approves First Fully Driverless Buses for Public Roads
Norwegian regulators have granted permission for fully driverless, level‑4 buses operated by state‑owned Vy and regional authority Kolumbus to run on public streets. The e‑ATAK electric buses, built by Turkey’s Karsan and equipped with Adastec’s autonomous system, will begin a...
AI Advances in Necrotizing Enterocolitis: Challenges Ahead
Artificial intelligence is being explored to predict and manage necrotizing enterocolitis (NEC), a rare but lethal condition in pre‑term infants. Researchers report that limited case numbers and heterogeneous data cause overfitting, while most models lack multicenter external validation. Low positive...
New AI Chatbot Uses Medical Protocols to Guide Patient Care Decisions.
UC San Diego researchers unveiled a multi‑agent AI chatbot that uses American Medical Association flowcharts to guide self‑triage. The system matches patient symptoms to protocol‑based questions, translating clinical language into lay terms. In over 30,000 simulated dialogues it chose the...
Yale Duo Secures $5.1M Pre‑Seed to Launch AI‑Powered iMessage Social Network
Yale seniors Nathaneo Johnson and Sean Hargrow closed a $5.1 million pre‑seed round for Series, an AI‑enabled social network that lives entirely inside iMessage. Backed by Venmo co‑founder Iqram Magdon‑Ismail, Pear VC, Reddit CEO Steve Huffman and GPTZero founder Edward Tian, the funding...
Dairy Queen’s AI‑Only Drive‑Throughs Prompt Customer Backlash
Dairy Queen is rolling out drive‑through lanes that operate without human staff, relying solely on artificial‑intelligence ordering systems. The move has provoked backlash from customers who find the experience frustrating, underscoring the tension between efficiency drives and consumer expectations in...
Meta Launches AI Business Assistant Worldwide for Advertisers
Meta has opened its AI Business Assistant to advertisers of all sizes in major global markets, adding local‑language support and embedding the tool directly in Ads Manager, Meta Business Suite and Business Support Home. The rollout promises faster campaign insights,...
Swiss Researchers Show Robots Learning Complex Tasks by Watching Humans
A team at École Polytechnique Fédérale de Lausanne published a Science Robotics paper showing robots that learn complex tasks by observing humans. The work, described as a breakthrough, could reshape industrial automation and raise ethical questions about robot autonomy.
AI Skills Tests Replace Misleading Resumes in Hiring
The Resume Had a Good Run HR leaders are retiring the traditional CV as the first impression in hiring, replacing it with AI-driven skills assessments that test actual capabilities. SHRM's 2026 HR trends data shows 70% of candidates admit to misrepresenting...
Free Software Foundation Says 'Responsible AI' Licenses Which Restrict Harmful Uses Are Unethical and Nonfree
The Free Software Foundation (FSF) announced that "Responsible AI" licenses (RAIL) are both nonfree and unethical. RAIL attempts to block AI and machine‑learning software from being used in listed harmful applications such as surveillance and crime‑prediction, yet it fails to...
50+ Ready-To-Use Claude Prompts: Turn Claude Into Your Content Machine
Sifu Yik’s Substack post shares 52 ready‑to‑use Claude prompts that span strategy, writing, social media, SEO, video, analytics and brand positioning. The guide stresses treating Claude as a collaborative team member by providing clear context, goals, format specifications and iterative...
DeepSeek V4 Boosts Open‑Source AI, Cuts Nvidia Dependence
Three reasons why DeepSeek’s new model matters The long-awaited V4 is more efficient and a win for Chinese chipmakers. 1. It breaks new ground for an open-source model. 2. It delivers on a new approach to memory efficiency. 3. It marks...
GStack v1.13 Enables Cross‑Modal Synthesis with GPT‑5.5
GStack v1.13 shipped - simple feature, I'm spending a lot more time using GPT-5.5 with Conductor and so I realized I sometimes wanted to get Opus 4.7's take on a change before I kept going. Cross-modal synthesis is something I...
Exploding Model Capabilities Chart Is Key Benchmark
NEW EPISODE: Understanding the most viral chart in AI @tracyalloway and I talk to @ChrisPainterYup and @joel_bkr of @METR_Evals about what this chart of exploding model capabilities is actually showing, and why they think it's the key benchmark to watch https://t.co/11P3UKZkE8 https://t.co/Dt3HYNOWyu
Facial Age Acceleration Predicts Earlier Death, Varies by Job
Face photo-based age acceleration predicts all-cause mortality and differs among occupations "We found that face photo-based age predicts all-cause mortality for middle-aged and older individuals meaning that those age faster based on their face photo die sooner." https://t.co/QUUzK4fe9G
AI Success Depends on Process, Not Magic
AI isn’t magic. It’s a process. 8 steps: 👉define problem 👉collect/prepare data 👉choose model 👉train 👉evaluate 👉fine-tune 👉deploy 👉ensure ethics & safety Real value comes from running this loop well. #AI #MachineLearning #DataScience #ResponsibleAI https://t.co/9dCrDBVWrW
CTO Prays to Dario Before CEO's First AI Demo
CTO praying to Dario before CEO gives first live demo of the AI agent to an enterprise customer
Virtual Assistants Enable Deeper Financial Engagement, Not Just Automation
The opportunity is not just automation. It’s deeper engagement. Virtual assistants can help customers manage money, not just answer questions. https://t.co/dcRagBD5q7
AGI's Toughest Challenge: Matching Socks, Not Math
The first self-aware robots are coming for your laundry. Turns out the hard part of AGI was never the math. It was matching socks.
AI Chat Models Are Just Prompts, Not Builders
ChatGPT is just a less useful version of Codex that can't build and ship stuff. Same for regular Claude vs. Claude Code. It's all just prompting the AI at the end of the day.
AI Imagines Stata Center's Futuristic 2500 Design
What does AI think the Stata Center will look like in the year 2500? ChatGPT, Meta AI, & Grok chimed in. https://t.co/dhIJxYKs2w