Today's AI Pulse
Anthropic shuts down Claude Fable 5 and Mythos 5 after U.S. export‑control warning
Anthropic disabled access to its Claude Fable 5 and Claude Mythos 5 models after the U.S. Commerce Department warned that foreign nationals could exploit the systems. The warning followed alerts from Amazon and other tech firms about a potential jailbreak that could facilitate cyber‑security attacks. Anthropic declined the administration’s request for a voluntary shutdown.
Also developing:
By the numbers: Meta scraps $2B Manus acquisition amid China divestiture order
Navitas Vs. Poet: Which AI Infrastructure Stock Is the Better Buy?
Navitas Semiconductor (NVTS) and Poet Technologies (POET) are both riding the AI‑infrastructure surge, but their trajectories differ sharply. Navitas has secured multiple tier‑1 AI‑chip contracts, giving it solid validation and a more mature revenue base. Poet, a smaller player, relies on the rapid scaling of its AI‑accelerator platform, offering investors a high‑growth, high‑risk proposition. Both stocks surged dramatically in 2026, prompting a debate over proven performance versus speculative upside.
Anthropic and OpenAI Now Capture 89% of Top AI Startup Revenue, with 34 Leading Startups Generating $80B in Annualized Sales
Anthropic and OpenAI now dominate the AI startup landscape, capturing roughly 89% of revenue across the 34 leading generative‑AI firms. Together these companies generate about $80 billion in annualized sales—$6.6 billion per month—a 112% increase from six months earlier. Anthropic has overtaken...
AI Accelerates Finance Work, Threatening White‑Collar Jobs
“in the last few months, there has been a step-change function in the productivity of the AI toolkit … work that we would usually do with people with Master’s and PhDs in finance over the course of weeks or months being done...
A Blueprint for Scaling Recommender Systems
Meta unveiled a two‑tier architecture for hyperscale recommender systems that separates a massive Foundation Model (FM) from lightweight surface‑specific Expert models. The FM learns universal, lifelong user representations and generates target‑aware embeddings that capture a user’s interest in each candidate...
The Smart Money Isn't on Anthropic — It's on the Patterns They Keep Setting
Anthropic briefly overtook OpenAI in business adoption, capturing 34.4% versus OpenAI's 32.3%. Hours later, Sam Altman pledged two months of free Codex to any firm that disabled Claude Code, prompting Anthropic to boost Claude Code limits by 50% through July....
Inside How Anthropic Is Building the Next Claude | Alex Albert
Anthropic’s research PM Alex Albert gave an inside look at how the company builds the next generation of Claude, its flagship large language model. The team defines high‑impact capabilities—such as coding and knowledge work—before pre‑training, using direct feedback from enterprise...
What Data Agent Benchmarks Do and Don't Tell Us
The recent AI Council conference highlighted a universal pivot toward AI infrastructure, with companies ranging from startups to established players redefining themselves as providers of context, orchestration, or compute layers for agents. New AI‑native databases such as LanceDB are being...
Founders Boost Output with Anthropic's Claude Cowork AI Hacks
Entrepreneurial leaders are integrating Anthropic’s Claude Cowork AI agent into daily workflows, automating spreadsheet cleanup, slide creation and email handling. The tool’s built‑in virtual machine lets users delegate routine work without installing external developer tools, promising a new efficiency frontier...
Elon Musk Pledges up to $119 B for Massive AI Chip Fab, Sparking Industry Debate
Elon Musk announced plans to pour up to $119 billion into a vertically integrated chip‑making operation, dubbed “Terafab,” to supply AI compute for Tesla, SpaceX and xAI. The proposal pits Musk’s ambition against entrenched players like TSMC and raises questions about...
OpenAI’s $4 B Deployment Company Triggers 700% Surge in Forward‑Deployed Engineer Demand
OpenAI unveiled a $4 billion Deployment Company to staff enterprises with forward‑deployed engineers, while Google Cloud announced hundreds of openings for the same role. The combined hiring push has lifted demand for forward‑deployed engineers by more than 700% in the past...
Growth Catalyst Group Hires Maia Benson as CCO to Steer AI‑powered Sales Expansion
Growth Catalyst Group of Companies announced the appointment of Maia Benson as chief commercial officer. Benson, who helped build Shopify's fulfillment network into a multi‑billion‑dollar operation, will lead GCG's go‑to‑market and AI‑driven growth strategy across its supply‑chain portfolio.
Anthropic Pulls Away, OpenAI Strikes Back, and Google's Gemini Rising
Anthropic reported an 80‑fold year‑on‑year revenue surge, pushing its annualized run rate above $44 billion and overtaking OpenAI in verified enterprise customers. The company also sealed a $200 million, four‑year partnership with the Gates Foundation to support global health and education initiatives....
Hidden Audio Can Hijack AI Without User Awareness
New research reveals how nearly imperceptible audio in videos, calls, or music can trigger unauthorized AI actions without users noticing. https://spectrum.ieee.org/voice-ai-audio-attacks?share_id=9501042
Evolve SEO Into Predictive, AI‑First Content Strategy
Is your Marketing Search Strategy Built for an AI-driven world? As generative AI transforms how users discover and evaluate information, SEO must evolve from a channel strategy to a predictive, full-funnel content system. At Marketing (re)Focus 2026 hosted by Grazitti...
Voice AI Systems Are Vulnerable to Hidden Audio Attacks
Researchers unveiled AudioHijack, an adversarial technique that embeds inaudible commands in audio clips to hijack large audio‑language models (LALMs). Tested on 13 open and commercial models—including Microsoft’s service—the method achieved 79‑96% success, forcing models to conduct web searches, download files,...
Google-Agent: The Web’s New Visitor Just Got An Identity via @Sejournal, @Slobodanmanic
On March 20, 2026 Google added Google‑Agent to its official list of web fetchers. Unlike Googlebot, the agent only runs when a human asks an AI assistant—such as Project Mariner—to browse a page, acting as the user’s proxy. Google classifies it as a...
Agentic Coding: Keep Control, Don’t Let AI Drive
Agentic coding isn't vibe coding with a better name. People use these two terms interchangeably, but they are not the same thing. Vibe coding is when you let Claude or Cursor write code for you and you ship it without reading a...
AI Agents Lack Trust in High‑Risk Industries Without Domain Data
AI agents face trust issues in ‘high-risk’ industrial sectors: experts Expert says large language models struggle in professional domains due to lack of industrial data and knowledge during training https://t.co/FVaDeQ9UqG via @scmpnews
AI Backlash Becomes a Real Business Risk
Public backlash against artificial intelligence is intensifying, with recent polls showing a majority of Americans—over 70%—believe AI is moving too fast. Only 18% of Gen Z feel hopeful about the technology, while negative sentiment has risen from 34% to just...
Why Advisors Are Betting on AI Infrastructure Over Apps
Financial advisors are shifting focus from AI applications to the underlying infrastructure that powers them. Stocks tied to AI hardware, data centers and storage have posted gains of 200%‑plus this year, with SanDisk up 464% and DigitalOcean up 240%. Advisors...
AI Tools Are Rewriting Business Security and Not in a Good Way
Founders are rapidly adopting AI tools, but the speed of deployment is outpacing security safeguards, leading to data leaks. A recent breach at Vercel occurred when an employee linked a third‑party AI app to a corporate Google account, exposing database...
The Secret to Self-Hosting OpenClaw Without the Headache
Self‑hosting OpenClaw via EasyPanel on a virtual private server (VPS) offers a structured, low‑maintenance way to run AI agents while retaining full control over the underlying infrastructure. The EasyPanel dashboard streamlines installation of OpenClaw and ancillary services such as Postgres,...
Next Token Prediction Is a Misleading Term
The article argues that labeling large language models (LLMs) merely as “next token predictors” is misleading. While pre‑training does involve predicting the next token, the process forces models to learn grammar, factual knowledge, and logical patterns across very long contexts....
By 2047, Karnataka Plans to Use AI to Run Power Grid Operations
Karnataka's energy department announced a roadmap to fully automate its power grid operations with artificial intelligence by 2047. The plan promises real‑time monitoring of assets, distribution lines and transformers, cutting response times from weeks to minutes. Officials highlighted the creation...
‘The Train Has Left the Station’: Workers Are Cashing in by Teaching AI to Do Their Jobs — some Earn...
Workers are turning to high‑pay side gigs training artificial‑intelligence systems, with platforms like Mercor paying experts up to $350 an hour. Hollywood writer Ruth Fowler and doctors such as those in Mercor’s Physician Talent Network illustrate how professionals across fields...
Anthropic Rolls Out Claude for Legal with 20+ Connectors and 12 Practice Plugins
Anthropic introduced Claude for Legal, adding more than 20 Model Context Protocol connectors and 12 practice‑area plugins, while Freshfields signed a multi‑year agreement to deploy the tool across its 33 offices. The launch targets the 87% of general counsel now...
AI‑Built Platform Saves Aging Parent’s Life, Sparks Home‑Care Tech Wave
Pratik Desai built an AI‑driven health assistant that warned of a life‑threatening complication in his mother’s cancer care, extending her life by 76 days. The success has ignited a surge of “vibe‑coded” tools for dementia and chronic‑illness caregivers, with early...
Mosaic Raises $3.8M to Launch Ultra‑Low‑Power Perception Chip for Smart Glasses
Swiss semiconductor startup Mosaic announced a new perception chip that provides real‑time object recognition and scene understanding for smart glasses while consuming a fraction of the power of traditional GPU‑based solutions. The company also closed a $3.8 million funding round to...
Can ELK Be Brute-Forced? Intertheoretic Reduction
The post asks whether Eliciting Latent Knowledge (ELK) can be solved by brute‑force intertheoretic reduction, i.e., proving that an AI predictor’s internal model reduces to a formal human physics model given unlimited compute. It outlines a two‑step plan: first formalize...
Priority Software Unveils AI‑Powered ERP V26.0 with aiERP Companion and Specialized Agents
Priority Software rolled out Priority ERP V26.0, embedding an aiERP Companion and a suite of AI agents that automate finance, sales and supply‑chain tasks via natural‑language prompts. The upgrade aims to cut manual effort, improve decision quality and scale with...
GitHub Unveils Standalone Copilot Desktop App to Centralize AI Coding Agents
GitHub, a Microsoft subsidiary, released a technical preview of a standalone Copilot desktop application that consolidates coding agents, issue tracking and pull‑request management. The app is available on macOS, Windows and Linux for Business and Enterprise customers, with Pro users...
Lotte Hotel Seoul Records Staff Motions to Train AI Robots for Banquet Service
Lotte Hotel Seoul is wiring its banquet staff to capture hand motions, creating a data set that will teach future robots to fold napkins and set tables. The pilot involves about ten employees and ties into South Korea’s broader $33 million...
Broadcom Becomes Core AI Chip Supplier for Google, Anthropic, OpenAI and Meta
Broadcom's custom AI chips now power the compute engines behind Google, Anthropic, OpenAI and Meta, driving a 106% year‑over‑year surge in AI sales and fueling a $100 billion revenue target for 2027. The move reshapes the hardware foundation of big‑data analytics...
OCR Studio Expands KYC Fraud Detection for AI-Generated Identity Documents
OCR Studio has upgraded its document‑verification platform with an anti‑fraud engine that can spot AI‑generated and morphed identity documents, including forgeries created by tools such as ChatGPT, Midjourney, Grok and NanoBanana. The technology examines the underlying image structure to detect...
Why Treating ChatGPT Like a Search Engine Is a Major Mistake In 2026
The article warns that using ChatGPT and similar models as a search engine undermines their strengths. It explains that advanced 2026 models like ChatGPT 5.2 and Gemini 3 process text, images, audio, and video, but require precise, context‑rich prompts to deliver useful...
Why Ideogram Stands Out in the AI Image Boom
Ideogram is an AI image‑generation platform that differentiates itself with a suite of user‑friendly features such as Magic Prompt auto‑refinement, four image choices per query, and precise text rendering. The service offers a free tier with limited daily credits and...
Greg Brockman Consolidates OpenAI's Product Teams to Build an "Agentic Future"
OpenAI has reorganized its product leadership, appointing co‑founder Greg Brockman as head of product strategy after interim coverage while AGI Deployment CEO Fidji Simo recovers from medical leave. Brockman announced a consolidation of product teams to focus on an “agentic...
AI Could Derail Everything From Global Financial Systems to Online Privacy: Would You Be Vulnerable to an Attack?
Anthropic’s new AI model Mythos can autonomously discover thousands of software vulnerabilities, prompting warnings from JPMorgan CEO Jamie Dimon that the financial system could become a target. A recent accidental leak of Claude’s source code highlighted how AI tools might...
OpenAI Introduces Personal Finance Assistant for Account Queries
OpenAI is launching a personal finance experience so you can connect up accounts and ask questions about your money. https://t.co/Vhg1tF4xRy
How Amazon Went From an AI Also-Ran to a Real Contender
Amazon has transformed its AI posture from a peripheral player to a serious competitor by marshaling roughly $200 billion in AI‑related spending, unveiling custom silicon, and striking strategic partnerships. AWS now offers a portfolio that rivals Microsoft’s Azure OpenAI services, with...
How AI Is Rewriting the Rules of African Retail
The McKinsey‑ICSC report finds AI reshaping retail, with agentic commerce projected to generate $1 trillion in U.S. B2C sales by 2030 and $3‑5 trillion globally. In Africa, a young, mobile‑first population and rapid urbanisation create fertile ground for AI‑driven convenience and experiential...
New Math Benchmark Reveals AI Models Confidently Solve Problems that Have No Solution
A 64‑researcher consortium released SOOHAK, a 439‑question math benchmark that separates graduate‑level challenges from 99 deliberately unsolvable problems. Leading models such as Google Gemini 3 Pro and GPT‑5 hit only 30% and 26% accuracy on the challenge set, while none surpass 50%...
AI's Future Hinges on Responsible Local Infrastructure
AI infrastructure is becoming the next big local political fight. Data centers power the AI boom, but they also compete for electricity, water and community trust. The AI winners will not only be those with the best models, but those who build...
Atlas Robot Learns Human-Like Moves via Reinforcement Learning
Reinforcement-Learned Atlas Performs Moves Inspired by Human Mocap and Animation via @ZappyZappy7 #Robotics #Engineering #ArtificialIntelligence #Innovation #Technology https://t.co/Udic9QKJTI
Artificial Intelligence and Legal Alignment
Jack Boeglin, a Penn Law scholar, argues that “legal alignment” can complement technical AI alignment by grounding system behavior in established legal norms. He warns that divergent regulatory regimes worldwide, especially the EU’s aggressive AI rules, force developers to either...
🔮 Exponential View #574: Inside Anthropic’s Rocket Ship; AI Pluralism; Love Commoditized, Context-Maxxing & Voltaire++
Anthropic’s CFO Krishna Rao revealed that enterprise customers have increased their AI spend fivefold over the past year, propelling the company’s annualized revenue from $250 million to roughly $50 billion. The firm now leads OpenAI in business adoption, according to Ramp, and...
US AI Leadership Hinges on Power and Compute Infrastructure
America cannot lead the AI race without building the power and computing infrastructure AI requires. On that I agree completely. But the piece gets the diagnosis wrong — and a wrong diagnosis produces the wrong cure. 🧵
Freshworks Launches Freddy AI Agent Studio to Automate Enterprise IT, HR, Finance
Freshworks announced Freddy AI Agent Studio at its Refresh 2026 conference, offering a no‑code environment to create custom AI agents for IT, HR and finance workflows. The platform promises to cut deployment cycles from quarters to weeks and addresses a...
Salesforce Commits $300 Million to Anthropic Tokens, Rolls Out Agentic AI Across Sales Workflows
Salesforce announced a $300 million spend on Anthropic tokens for 2026 and began deploying agentic AI across its core sales, service and marketing workflows. The move ties the CRM giant’s AI strategy to Anthropic’s Claude models and expands AI‑driven coding inside...
Amazon Unveils AI‑Powered Shopping Ads Tied to Streaming TV, Reaching 90% of U.S. Households
Amazon announced a new ad offering that blends AI‑driven campaign tools, authenticated audience data and premium streaming inventory. The platform claims to reach roughly 90% of U.S. households and will power “agentic shopping” experiences across live sports and on‑demand video.