Today's AI Pulse
Anthropic shuts down flagship models after U.S. export‑control order
Anthropic abruptly disabled its Fable 5 and Mythos 5 models to comply with a U.S. export‑control directive, highlighting the fragility of businesses that rely on closed‑source AI services. The shutdown triggered a sharp rally in Chinese open‑source offerings such as MiniMax and Zhipu as investors bet on downloadable models that can run on‑premise.
Also developing:
By the numbers: SpaceX to acquire Cursor in $60B all‑stock deal
Decart’s New World Model Can Simulate Hours of Photorealistic Driving — with some Caveats
Decart unveiled Oasis 3, a real‑time photorealistic driving world model accessible via API, targeting autonomous‑vehicle developers and broader physical‑AI use cases. The service costs $0.02 per second, with enterprise pricing tailored to specific workloads. The launch follows a $300 million funding round that lifted Decart’s valuation to roughly $4 billion, backed by investors such as Toyota, Adobe, eBay and Nvidia. While Oasis 3 delivers striking visual fidelity, reviewers note rapid scene degradation and limited physics realism over longer simulations.
Apple Bets on AI Overhaul at WWDC 2026 as Tim Cook Prepares to Hand Over the Reins
At WWDC 2026, Apple unveiled a sweeping overhaul of Siri and rolled out AI‑enhanced features across its core apps, built in part on Google’s Gemini models. The new Siri will operate as a standalone app offering conversational and visual intelligence...
Where AI Is Really Taking Jobs
The latest Signal Pro post examines real‑world data on AI’s impact on employment, citing the World Economic Forum’s 2025 Future of Jobs Report and Stanford’s Digital Economy Lab study. While the WEF forecasts a net gain of 78 million jobs by...
Los Alamos Method Helps Expose Hallucinations in Vision-Language AI
Los Alamos National Laboratory unveiled the Prelim Attention Score (PAS), a real‑time metric that detects hallucinations in vision‑language AI models. PAS monitors token‑level attention to determine whether outputs are grounded in the image or driven by prior generated text, achieving...
Disaggregating AI Compute to Break the Tokens Barrier
The AI compute market is hitting a token ceiling as demand outstrips supply, prompting heavyweights like Anthropic to pay $1.25 B per month for SpaceX’s Colossus capacity. This shift follows the “token‑maxxing” surge that drove hyperscalers to pour billions into massive...
Biohub: The Future of Biology Is Open-Source with Co-Founders Mark Zuckerberg, Priscilla Chan, and Head of Science Alex Rives
In this episode, Mark Zuckerberg, Priscilla Chan, and Alex Rives discuss the mission of Biohub and its $500 million Virtual Biology Initiative, which aims to accelerate scientific discovery by creating open‑source tools and massive biological datasets. They explain how Biohub combines...
Anthropic Warns LLMs Can Crank Out N-Day Exploits Cheap and Fast
Anthropic’s Mythos preview model demonstrated that large language models can create functional N‑day exploits in hours, a task that usually takes weeks of expert hacking. In a red‑team test of 21 known bugs, including 18 from Microsoft, Mythos generated exploits...
Essential Books to Become a 10x AI Engineer
10 Books that will make you a 10x AI engineer: 1 Building LLMs for Production 2 AI Engineering 3 Designing Machine Learning Systems 4 Build a Large Language Model 5 Designing Data-Intensive Applications 6 LLM Engineer's Handbook 7 Deep Learning 8 Hands-On Machine Learning with Scikit-Learn, Keras, and...
Poetic Emerges From Stealth with $50M From OpenAI to Automate Insurance Underwriting and Compliance
Poetic, an AI startup that operated in stealth, raised $50 million in a Series A round that values the company at $500 million, with investors including OpenAI, Founders Fund and Kleiner Perkins. The platform ingests a firm’s procedures, training videos and expert feedback...
Google Embeds NotebookLM Into Gemini
Google has rolled out Notebooks in Gemini, embedding its NotebookLM research engine directly into the Gemini AI chat app. The integration lets users create dedicated notebooks that store sources, files, and conversation history, with changes syncing instantly between Gemini and...
The AI Threat We Should Actually Be Talking About
Artificial intelligence firms such as Anthropic and OpenAI are racing toward IPOs that could value them at more than $1 trillion. While many call for slowing or pausing frontier AI development, the article argues the greatest danger lies in the concentration...
OpenAI Weighs Nvidia-Backed Lease for 10 GW Ohio Data Center Campus
OpenAI is in advanced talks to lease a proposed 10‑gigawatt data‑center campus in southern Ohio, a project that could cost at least $500 billion. Under a 20‑year lease, OpenAI would control the compute equipment while Nvidia would supply the hardware and...
Google to Pay SpaceX $30 Billion for AI Compute
Google has agreed to pay SpaceX $920 million each month for access to roughly 110,000 Nvidia GPUs, CPUs, memory and related components. The deal kicks off in October 2024 and runs through June 2029, delivering more than $30 billion in revenue if...
CoreWeave and Meta Announce $21 Billion Expanded AI Infrastructure Agreement
CoreWeave and Meta Platforms have signed an expanded, long‑term AI cloud agreement worth roughly $21 billion, extending through December 2032. The deal secures dedicated multi‑regional capacity for Meta’s inference workloads and incorporates early deployments of Nvidia’s Vera Rubin GPU platform. It builds...
Musk: 'Magic' Not Needed for SpaceX's Orbital AI Data Center Plan
Elon Musk outlined SpaceX’s plan to launch up to one million orbital AI data‑center satellites, leveraging Starlink‑derived solar panels, radiators and compute modules. The first‑generation AI1 satellite will be 20 m tall, 70 m wingspan, deliver up to 150 kW peak power and...
Google and FANUC Want to Build the First Truly Intelligent Factory Robots
Google and FANUC have launched a “Physical AI” partnership that embeds Google’s Gemini Enterprise large‑language model into FANUC’s more than 1.1 million industrial robots. The joint system combines multimodal AI, computer vision and sensor fusion to let robots understand natural‑language instructions...
Germany's National Security Council Greenights an AI Safety Institute Modeled After the UK's AISI
Germany’s National Security Council approved the creation of the German AI Security Institute (DE‑AISI), modeled after Britain’s AI Safety Institute. The new body will assess advanced AI models for cybersecurity risks and coordinate standards with international partners. DE‑AISI aims to...
One Prompt, One Platform: How Intuit Is Rethinking the Financial Journey with Agentic AI
Intuit unveiled an agentic AI platform that lets small‑business owners issue a single natural‑language prompt and receive a consolidated financial recommendation. The system spawns multiple specialized agents to pull cash‑flow data, credit history, payroll runway, and even pre‑approve a loan...
The 3-Min Read: Why Anthropic Is Becoming AI’s Reference Point
Anthropic closed a $65 billion Series H round on May 28, 2026, valuing the company at $965 billion, briefly making it the world’s most valuable AI startup. The firm now reports an annualized revenue run‑rate above $47 billion, driven largely by enterprise adoption of its...
Bedrock & Advertible Close the Agentic Loop From Decisioning to Creative
Bedrock Platform has partnered with creative‑infrastructure firm Advertible to launch the Creative Agent, an AI‑driven tool that generates and hosts ad creatives directly within Bedrock’s programmatic workflow. The solution uses Advertible’s sizeless rendering to transform a simple text prompt and...
Tractor Supply Taps AI to Help Scale Last-Mile Delivery Network
Tractor Supply is using artificial intelligence to scale its private last‑mile delivery fleet, which it began expanding in early 2025. The retailer shifted route‑building from drivers to territory managers equipped with AI tools, allowing drivers to concentrate on customer interaction....
Energy‑saving Tweak Trims LLM Training by 14%
LLM training burned 50 GWh for GPT-4, but researchers are developing ways to cut energy usage. This tweak could trim energy by up to 14 percent without affecting performance. https://spectrum.ieee.org/llm-training-energy-saving-trick?share_id=9593619
Timing Trick Cuts Energy Used in LLM Training by Up to 14 Percent
Researchers at the University of Twente demonstrated that adjusting GPU clock frequencies during large‑language‑model training can reduce energy consumption by up to 14 percent while adding only a 0.6 percent slowdown. The technique leverages dynamic voltage‑frequency scaling (DVFS) at the granularity of...
GitHub Copilot Is Generating More Code than Your Team Can Review: Why Senior Engineers Are Now the Bottleneck
GitHub Copilot enables developers to generate code at unprecedented speed, dramatically increasing raw output. However, organizations quickly encounter a new bottleneck: peer review and validation, as senior engineers become overloaded reviewing AI‑generated pull requests. The surge in code volume inflates...
AI Is Reaching Finance’s Core Systems: Here’s What It Takes to Run It There
AI adoption in finance remains limited, with only about 10% of firms deploying production‑grade models in core systems. Legacy architectures, built through years of regulation and acquisitions, isolate AI from trade capture, risk, and surveillance functions. Vendors propose a virtualized...
Broadcast Media Regulator Issues Guidelines on AI-Generated Content
Taiwan's National Communications Commission (NCC) issued administrative guidelines requiring broadcast outlets to clearly disclose any AI‑generated content used in news reporting. Aligned with the AI Act that took effect on Jan. 14, the rules demand prominent on‑air disclosures, attribution of third‑party...
The Sequence AI of the Week #875: Why Your Language Model Needs a Nap
The blog spotlights a new paper arguing that large language models suffer from anterograde amnesia after pre‑training, forgetting new information once a session ends. The authors propose a biologically inspired “sleep” phase that consolidates recent context into long‑term weights. Experiments...
6 Best AI Pricing Platforms for Omnichannel Retailers in 2026
The article ranks six AI‑driven pricing platforms tailored for omnichannel retailers, naming Quicklizard as the top choice for its blend of demand forecasting, inventory‑aware repricing, competitor intelligence, and real‑time price synchronization. Competera, Omnia Retail, Revionics, Wiser, and 7Learnings round out...
.jpg?width=1280&auto=webp&quality=80&disable=upscale)
How a Ghanaian Side Hustle Became a Webby-Winning AI Success Story
Ghanaian‑born entrepreneur Dominic Damoah turned a side‑hustle into Human in the Loop (HITL), a platform that adds human oversight to AI‑driven workflows. HITL earned the 2026 Webby People’s Voice Award for Best Responsible AI Implementation, highlighting its safety‑first design. The...
Grasshopper Launches AI-Based Treasury Investment Service
Grasshopper Bank has launched an AI‑powered treasury management service in partnership with corporate robo‑adviser Waldo, giving its business‑banking clients automated investment recommendations. The offering, available via a waitlist for enterprise users, requires a $250,000 minimum balance to access the treasury...
AI Diagnoses Brain Tumors in Minutes Rather Than Weeks
Researchers in Heidelberg unveiled Hetairos, an AI system that reads standard H&E‑stained slides to classify over 100 molecular subtypes of central nervous system tumors. Trained on more than 11,000 images from 9,600 patients, the model reaches 68% overall accuracy and...
Motive Brings AI Coach to the UK: Organisations Can Deliver Personalised Driver Coaching Automatically with Custom Avatars
Motive has launched AI Coach in the United Kingdom, an AI‑powered avatar system that automatically creates personalized video feedback for drivers. The solution lets managers select pre‑built or custom‑recorded avatars that mimic a familiar face and voice, delivering weekly coaching...
Fish Brothers Expands AI Use After 492% ROI Claim
Fish Brothers Group, which runs 14 UK dealerships, has rolled out ELLA’s optimAIze AI platform across its sales and after‑sales operations after reporting a 492% return on investment in Q1 2026. The system engaged 83% of customers who bought a...
Anthropic Unveils Claude Fable 5, First Public Mythos-Class AI Model
Anthropic announced Claude Fable 5, the first generally available Mythos‑class AI model, pricing it at $10 per million input tokens and $50 per million output tokens. The launch pairs frontier performance with extensive safeguards that block high‑risk requests in cybersecurity,...
Why Your Most AI-Savvy Employees Are Driving Shadow AI
Employees with deep generative‑AI knowledge are increasingly sidestepping corporate restrictions, creating a shadow AI ecosystem. A LexisNexis study shows 74% of AI‑trained workers use unauthorized tools, compared with just 17% of those without training. CIOs face a paradox: training fuels...
The Missing Link in Retail AI ROI: Connected Process Chains
Retail executives are pouring billions into AI, yet 96% report no measurable ROI because most deployments remain isolated point solutions. The article argues that true value emerges only when AI is woven into end‑to‑end process chains, linking forecasting, inventory, logistics...
The Lean AI Plan for Action at VietBank
VietBank’s CIO, Nghia Tran, has built a lean, in‑house AI engine that powers a Smart Office Tracking system and an Intelligent Management System using a self‑hosted LLM. The AI suite cut document approval cycles by 35% and earned a 2025...
AI Is Becoming a Single Point of Failure — and Most Companies Don’t See It
Artificial intelligence is rapidly becoming a core operating layer, but its reliance on constrained, vendor‑controlled infrastructure creates a hidden single point of failure. Recent shifts in Microsoft Copilot access and other tiered models illustrate that capacity is not unlimited. While...
Haut.AI Brings AI Skin Intelligence to VivaTech 2026
Haut.AI showcased its AI skin‑intelligence suite at VivaTech 2026 in Paris, inviting beauty brands, retailers and investors to its booth. Its Face Analysis 3.0 platform can evaluate more than 29 skin health parameters and over 150 facial biomarkers from a single...
Artificial Intelligence Sneaks Into the World Cup Thanks to Google Gemini
Google has partnered with Argentina’s national soccer team to make its Gemini AI the main global sponsor for the World Cup. The Gemini logo will appear on the team’s training kit, and the AI will be used to analyze plays,...
Batteries Plus Technology Leader Karthik Jambulingam Drives AI-Ready Enterprise Architecture Approach for Scalable AI Adoption.
Batteries Plus, the nation’s leading battery and power solutions retailer, has completed a multi‑year enterprise architecture overhaul that positions the firm for large‑scale AI adoption. Senior Director Karthik Jambulingam led the shift to a cloud‑native, data‑driven infrastructure, enabling faster software...
PromptBio Launched All-New Agentic AI Platform For Life Sciences, Turning Scientific Intent Into Executable Experiments
PromptBio, a California AI firm, unveiled an all‑new Agentic AI platform that lets scientists turn natural‑language questions into executable bioinformatics workflows. The system pairs a conversational interface with a proprietary Chief Scientific Orchestrator that coordinates multiple specialized agents for literature...
Generative Vision Interview Questions #3 - The KL Divergence Paradox
In a senior AI engineer interview at OpenAI, candidates are asked why the DDPM loss remains tractable despite a 1,000‑step diffusion trajectory. The answer hinges on the Gaussian Mirror Assumption, which forces the reverse transition to be an isotropic Gaussian....
VCs Can't Get Enough of Legal Startups. Andreessen Horowitz Just Invested in One Taking Work From Patent Lawyers.
Fearn, an AI‑driven patent‑drafting startup, closed a $5.5 million seed round led by Kindred Ventures with participation from Andreessen Horowitz and others. The platform lets inventors generate a complete patent application in minutes for a flat $2,000 fee, scoring quality and...
AI and PR: The Future of Media Engagement with Iskren Lilov, Muck Rack
In this episode of Marketing B2B Tech, global marketing manager Iskren Lilov explains how Muckrack’s AI‑powered communications platform combines media monitoring, journalist databases, press distribution and geo‑measurement to help PR teams demonstrate value. He highlights that 99% of AI citations...
Viridien Taps AI, Data Science and Domain Expertise for Evergreen Data Solution Launch
Viridien has introduced Evergreen Data Solution, an AI‑powered platform that transforms fragmented subsurface data into a unified, analytics‑ready foundation. The system automatically ingests, standardises and validates geoscience and seismic data across on‑premise, cloud and hybrid environments. By continuously updating the...
AI‑Native Ryt Bank Proves AI Banking Works
Ryt Bank launched in August 2025. Seven months later: 1.2 million users, 25 million transactions, monthly volumes 35x higher than at launch. That is not a perpetual AI pilot. That is a working AI-native bank. ↳ NPS score of 76 — exceptional...
Voice AI Revolutionizes Communication Across All Industries
Voice AI transforms #Communication in every industry, from #CustomerService to healthcare, education, and #Telecom 🙌 If you want to understand #VoiceAI technology, its impact & what's coming next, this guide I just published on Medium covers it all 🔗...
AI Red Teaming Comes of Age
AI red teaming, once a niche cybersecurity practice launched by Microsoft in 2019, has been reshaped by the rise of large language models like GPT‑4. Traditional deterministic testing proved ineffective, forcing teams to rebuild tools, methodologies, and even job definitions....
AI’s Agents’ Control Layer: What Separates Demos From AI That Deploys
The article argues that AI agents need a robust control layer—architectural guardrails, permissions, cost caps, approval gates, and audit trails—rather than relying on prompts. Demonstrations can be impressive, but without structural safeguards agents become liabilities. Trust gaps such as missing...