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
Magnet Forensics Unveils Magnet AI, Advancing The Next Era Of Digital Investigative Intelligence
Magnet Forensics announced Magnet AI, an intelligence engine that powers new AI‑driven features across its platform. The rollout includes Intelligent Search, which uses natural‑language queries to surface relevant media in seconds, and Intelligent Insights, which generates structured investigative summaries, suggested analysis paths, and citation‑verified findings. Both tools aim to cut manual data‑filtering time, reveal hidden connections, and keep final decision‑making in human hands. Early‑access to Intelligent Insights is now available in Magnet Review.
Making an AI a Multiplier Instead of a Threat
Boston Consulting Group’s experiment with 758 consultants showed AI users completed tasks 25% faster and delivered 40% higher‑quality output, with the lowest performers improving by 43%. The article argues that treating generative AI as a multiplier—not a threat—unlocks leverage across...
NTT DATA Unveils Multivendor Agentic Services Experience for Enterprise Infra
NTT DATA launched its Software Defined Infrastructure (SDI) Services Agent, a conversational, multivendor orchestration platform for enterprise IT. The agent acts as a digital twin, using natural‑language prompts to sense, reason and act across networking, hybrid data‑centers, cybersecurity and digital‑workplace...
EU Plans Competition Push For Cloud, AI
The European Commission announced that its Digital Markets Act (DMA) will increasingly target cloud and artificial‑intelligence services, building on progress made in other sectors. Gatekeepers such as Alphabet, Amazon, Apple, ByteDance, Meta, and Microsoft remain subject to stricter obligations, and...
If AI Can Model Cells, Science Can Deliver Cures
The Biohub Institute announced the Virtual Biology Initiative, a $100 million pledge to generate open‑source cellular data for AI training. Partnering with the Allen Institute, Broad Institute, NVIDIA, Wellcome Sanger and others, the effort aims to build massive, public datasets that...
With Nemotron 3 Nano Omni, Nvidia Reveals What Really Goes Into a Modern Multimodal Model
Nvidia unveiled Nemotron 3 Nano Omni, a 30‑billion‑parameter open‑source model that natively handles text, images, video, and audio. The hybrid Mamba‑Transformer with Mixture‑of‑Experts activates roughly three billion parameters per query and supports a 256,000‑token context window. Training spanned seven stages, processing...
From Encyclopedias to AI: How Knowledge Is Changing the Way We Work
The article traces the evolution of knowledge tools from static encyclopedias to AI‑driven copilots, highlighting how each stage reshapes work. In the AI era, systems not only retrieve data but also analyze, summarize, and generate content, turning users into directors...
Our AI Race with China: The Good News
The United States and China are locked in a fast‑moving AI rivalry, with Chinese models narrowing the performance gap and occasionally leveraging stolen U.S. technology. In April, the House Select Committee on the Chinese Communist Party held a hearing that...
LLM System Design Interview #36 - The Isomorphic MLP Trick
In a Meta senior AI‑engineer interview, candidates are asked to replace a ReLU‑based feed‑forward network with SwiGLU while keeping the classic 4× expansion factor. The trap is that SwiGLU introduces a third weight matrix, inflating the FFN parameter count by...
3 Ways You Didn’t Know AI Is Changing The Future of Work
Artificial intelligence is shifting from a productivity buzzword to a tangible cost center within enterprises. Companies now track AI compute spend alongside salaries because each prompt and inference adds a payroll‑like expense that can reach five‑figure annual totals for heavy...
How AI Could Help Combat Antibiotic Resistance
Antibiotic resistance now kills over a million people annually and could claim 40 million lives by 2050. Traditional culture‑based diagnostics take days, forcing physicians to guess treatments and driving misuse of antibiotics. AI‑powered diagnostics are achieving more than 99% accuracy and...
AWS Leans on Prior Ingenuity to Face Future AI and Quantum Threats
AWS marks its 20‑year anniversary while confronting AI‑driven attacks and the looming quantum‑computing risk. The Nitro hardware platform gives AWS a “zero‑human” infrastructure, enabling isolated bare‑metal instances and protecting encryption keys. Early adoption of symmetric encryption means most data at...
Can We Achieve “Super Abundance” Without AI Doom?
In a new episode of *What Could Go Right?*, author Sebastian Mallaby sits down with host Zachary Karabell to dissect the mindset of DeepMind founder Demis Hassabis, the architect of today’s most powerful AI systems. The conversation explores Hassabis’s “Ender’s...
BYD Hikes ADAS Price 21% as AI Demand Triggers DRAM Shortage
BYD announced a 21% price increase for its God’s Eye B ADAS, raising the option from roughly $1,450 to about $1,680 per vehicle starting in May. The hike targets select models across the Dynasty, Ocean, and Fang Cheng Bao lineups, while customers who...
Dex Lands $5.3M to Grow Its AI-Driven Talent Matching Platform
London‑based Dex, an AI‑driven recruitment platform for software engineers, closed a $5.3 million seed round led by Notion Capital with participation from Andreessen Horowitz’s Speedrun programme and other investors. The funding brings Dex’s total capital to $8.4 million and will be used...
Massive Bio Partners with OpenAI to Broaden Clinical Trial Access
Massive Bio has teamed up with OpenAI under the Impact Hours programme to automate clinical‑trial eligibility screening. The AI engine translates complex sponsor criteria into machine‑readable parameters, enabling real‑time, automated patient pre‑screening for oncology and haematology studies. The partnership includes...
Rise Launches AWS-Backed AI and Cloud Training Initiative
Rise has launched Rise Upskill, a global, AWS‑backed training program offering two foundational AWS certifications—Cloud Practitioner and AI Practitioner. The free, online courses run from June 1 to September 1 and accept up to 50 participants, with applications due May 25. Training is...
Gen AI Fuels Insurance Fraud Arms Race
Generative AI is dramatically amplifying insurance fraud, with AI‑enhanced cases soaring from under 20,000 in 2022 to more than 80,000 in 2025. Fraudsters now generate entire claim packages—photos, invoices, and assessments—that appear internally consistent and can evade traditional automated checks....
How AI Is Changing the World of Development Grantmaking
Rhodri Davies explores how artificial intelligence is reshaping development grantmaking, a topic featured in Devex. AI-driven analytics are automating eligibility screening, predicting project impact, and extracting insights from legacy grant data. Philanthropic foundations are adopting predictive models and real‑time dashboards...
Top AI-Powered Vendor Risk Management Platforms for SaaS Companies in 2026
Vendor risk breaches now account for roughly 30% of incidents, prompting regulators like Europe’s DORA and the SEC to tighten third‑party oversight. AI‑powered TPRM platforms promise continuous, machine‑speed visibility, turning multi‑page SOC 2 reports into concise insights and auto‑generating tickets in...
Infor in the AI Age - a Progress Report
Infor reported robust SaaS revenue growth and roughly 30% booking expansion among its larger customers, highlighting strong demand for its AI‑enhanced ERP suite. The company now offers a fully multi‑tenant, cloud‑native platform that embeds generative, agentic and algorithmic AI across...
Extra #9 - The Regression Playbook Part 1 (Code)
The blog post launches Part 1 of a five‑part Regression Playbook, demonstrating how to build, train, and visualize five core regression techniques using Python’s scikit‑learn library. It walks through Linear Regression, Stochastic Regression, Decision Tree Regression, Random Forest Regression, and k‑Nearest...
Your AI-Coding Budget Just Got a Lot More Complicated
AI‑coding tools have moved from simple chat interfaces to agentic systems that execute commands, call APIs, and access the internet, dramatically increasing token consumption. Vendors such as GitHub Copilot, OpenAI, Windsurf, and Cursor are responding with token‑based pricing, tighter limits,...
Zapier Survey: 77% of Enterprise Leaders Say AI Skills Are Urgent, but Most Companies Still Aren’t Training Their Workforce
Zapier’s AI Skills Crisis survey shows 94% of enterprises already use AI, yet 77% of leaders deem AI skill development urgent. Formal training reaches only about half of IT and engineering staff and under 40% of sales, marketing, HR, and...
Greece’s AI Smart Policing System Ruled Unlawful After €4 Million Public Spending\
In 2019 the Hellenic Police awarded a €4 million (≈ $4.3 million) contract to Intracom Telecom for a “Smart Policing” system that equips officers with portable devices for facial‑recognition, fingerprint and license‑plate scanning. The AI‑enabled tools were intended to speed up identity checks...
YOLO26-Pose Delivers 17‑keypoint Pose in 1.
YOLO26-Pose tracks 17 human keypoints in a single forward pass. Smallest variant: 1.8 ms on a T4 GPU. ⚡ → RLE for sharper localization → NMS-free inference (predictable latency) → MuSGD for stable training Full breakdown 👇 https://t.co/8OaxzdrCPx #ComputerVision #YOLO26 Optional thread version: 1/ YOLO26-Pose is here. It predicts...
AI Giants Tighten Grip, Raising Platform Lock‑In Risks
SAP, Salesforce, and Microsoft lead in AI layer strategies, but this comes with a risk of platform lock-in. According to analysts, they're becoming the 'better drug dealers' of the tech world, making it harder to switch away. #Tech #AI #LockIn...
Former Sugar Execs Launch Brief
Former SugarAI leaders Larry Augustin, Clint Oram and Zac Sprackett launched Brief, a software platform that creates an Executive Intelligence Layer to keep decision‑making context alive across meetings, emails, Slack and CRM. The system records commitments, rationales and relationship history,...
China Halts New Robotaxi Licenses After Baidu Failures
* @elonmusk's phone rings* China: "hey Elon, you know the thing we did where you came in as the first-ever solo foreign automaker to jumpstart our EV industry?" Elon: "yes I recall your welcome" China: "what if we did this again, but with......
AI Failures Reveal Data Quality as Critical Priority
Data Quality Assurance broke out this week. Not because anyone got excited about data quality. Because AI deployments started failing. The models were fine. The data wasn't. 18 months later, every serious team is circling back to the unsexy work. https://t.co/6W59KI6Zhb
The Smarter AI Gets, the Less You Can Trust It on the Hard Stuff
Anthropic’s new ICLR 2026 paper shows that as language models become larger and reason longer, their errors shift from systematic to random, a phenomenon the authors label “incoherence.” The study, which evaluated Claude Sonnet 4, OpenAI’s o3‑mini/o4‑mini and Qwen 3 across GPQA, MMLU,...
New Tools Enable Managing Hundreds of AI Agents
Tools to help workers + devs handle lots of subagents are just emerging. Prediction: We’ll all have to get skilled at guiding hundreds-thousands of agents soon. Watch: @OpenAI Swarm + Symphony @OpenSwarm_ @Kimi_Moonshot Agent Swarm @Swarms_corp
The EU AI Office Must Prioritise Setting up the Advisory Forum
A coalition of 35 civil‑society groups and researchers has urged the EU AI Office to publish a clear timeline for establishing the Advisory Forum, the sole formal channel for non‑governmental input on the AI Act. The Forum, whose call for...
SK Hynix Begins Mass Production of 192GB AI Server Memory
SK hynix has entered mass production of a 192GB SOCAMM2 memory module built on its sixth‑generation 10nm‑class LPDDR5X process. The module, aimed at AI servers, delivers more than double the bandwidth of conventional RDIMMs while cutting power use by over...
Third Space Learning Raises £4.4m to Scale AI Tutoring
Third Space Learning, the Swindon‑based online education platform, has raised £4.4 million (about $5.6 million) to accelerate its AI‑driven maths tutoring service, Skye. The round was led by the British Business Bank’s South West Investment Fund with participation from Maven Capital Partners,...
AI Documentation Tool Transforms Adult and Children’s Social Care
System C is rolling out its AI‑driven FormFlow Assistant to 15 English local authority social‑care teams, automating real‑time documentation during assessments. Early data show documentation time halved, with a 68% efficiency gain in the paperwork stage and a 40% overall...
Exploring Instability Risks in the U.S.-China AI Rivalry
The United States and China are accelerating a strategic contest to dominate artificial intelligence, raising concerns that either side could resort to cyber attacks or kinetic force to impede the other's progress toward artificial general intelligence. RAND’s research team has...
AI Can Make Software Now. That Changes Everything, with Paul Ford
In this episode, host Matt Buchel talks with technologist and writer Paul Ford about the rapid emergence of AI-powered code generation, which Ford calls "vibe coding." He explains how recent advances—particularly Anthropic's Cloud Code and improvements in large language models—have...
Mirror, Mirror… Are LLMs the Fairest of Them All?
Shirley Marschall’s column argues that large language models (LLMs) have become the default answer in ad‑tech, eclipsing more specialized machine‑learning tools. She highlights the mirror‑room effect where hype, mega‑deals and endless prompts create a self‑reinforcing narrative. The piece warns that...
SUNeVision Concludes Third Startup Program Cohort
SUNeVision Holdings, Hong Kong's largest data‑centre operator, completed the third edition of its Startup Programme, selecting four AI‑driven startups for awards. The cohort attracted nearly 100 high‑calibre applications spanning smart‑city, green tech, digital assets, EdTech and immersive entertainment. Winners—including maritime...
AI Token Costs Redefine Labor Budget for CIOs
1/ Here’s one of the first real attempts at bounding the spend envelope for AI token use per worker. And it changes how CIOs should think about labor entirely. This should not be seen as tooling cost. It’s the price of baseline productivity...
2Trust.AI and Carahsoft Partner to Scale AI Governance for Public Sector
2Trust.AI has partnered with Carahsoft Technology Corp. to distribute its AI governance platform to U.S. federal, state and local agencies. The solution will be available through Carahsoft’s major procurement vehicles such as SEWP V, ITES‑SW2 and NASPO ValuePoint. It offers...
A Writing Experiment Reveals AI's Likely Future: Individual Productivity Rises; Collective Intelligence Decays
A recent writing experiment shows AI‑assisted authors produce more creative individual stories, but those stories converge, indicating a loss of collective diversity. Researchers also demonstrate that language models trained on AI‑generated data experience knowledge collapse, erasing minority viewpoints and nuanced...
Former Startmate Boss Michael Batko Is Back in Founder Mode Building with Hourglass AI
Former Startmate CEO Michael Batko has launched Hourglass AI, an Australian firm that builds production‑ready AI systems on a fixed‑fee basis. The startup, co‑founded with 22‑year‑old Finlay Ekins, emerged from stealth in February and already serves clients such as a...
64‑Minute Cod
THIS GUY JUST DROPPED A 64-MINUTE CODEX MASTERCLASS 🤯 @rileybrown covers everything from GPT 5.5 browser agents to generating videos with Remotion and building reusable AI skills. Best Codex workshop I've seen online. https://t.co/xcDTJ4n5lb
How Coding Agents Become Legal Tech Allies
Vesence’s coding agents let lawyers describe a task in plain English and receive a custom script that manipulates PDFs, Excel files, Word documents, and Outlook emails on the spot. The agent runs in a sandboxed bash environment, writes code, executes...
DeepSeek‑V4 Preview Offers Million‑token Context, Low‑cost Inference
Recently, DeepSeek released a preview version of DeepSeek-V4 and simultaneously open-sourced its model weights. This marks a significant technological upgrade: two model variants, million-token context length, and ultra-low inference cost, which all draw strong industry attention https://t.co/nrEeCwxkxU
The Needle and the Damage Done
The post uses the century‑long evolution of recorded‑music formats—from wax cylinders to streaming—as a metaphor for today’s AI boom. It outlines how each new medium sparked novelty, legal gray zones, rapid obsolescence, and eventually consolidation into a commodity platform. The...
The Rise of AI-Native SaaS: Why Born-AI Companies Are Scaling Faster Than Ever
AI‑native, API‑first SaaS firms are outpacing traditional players, with several hitting half‑billion dollars in annual recurring revenue (ARR) by mid‑2025. The sector saw a 28% jump in M&A activity, closing 2,698 deals in 2025, making 2026 the record‑setting year for...
Met Police Federation Decries ‘Outrageous’ Palantir AI System
The Metropolitan Police Federation is threatening legal action over the force’s new Palantir AI system, which it says breaches officers’ privacy and GDPR rules. The pilot, launched last week, has already led to two arrests and two suspensions while hundreds...