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
Open-World Evaluations for Measuring Frontier AI Capabilities
The paper introduces “open‑world evaluations,” a new class of AI testing that places agents in messy, real‑world tasks rather than tidy benchmarks. It surveys ten recent experiments, outlines best practices, and launches the CRUX collaboration of 17 researchers to run such evaluations regularly. In CRUX’s first study, an AI agent built and published an iOS app to the Apple App Store, incurring roughly $1,000 in costs and making only two errors, one requiring manual fix. The authors argue these evaluations provide early warnings of emerging capabilities, such as automated app‑store spam.
Schneider Electric Unveils Next Generation Agentic Manufacturing Capabilities with Microsoft Azure AI at Hannover Messe 2026
Schneider Electric announced a next‑generation agentic manufacturing platform built on its EcoStruxure Automation Expert and Microsoft Azure AI at Hannover Messe 2026. The joint solution creates a single, traceable workflow that spans design, simulation, commissioning and operations across cloud and edge environments....
AI Mode Now Shows Sites Side‑by‑side for Seamless Research
Oh boy, I'm not sure publishers will be thrilled with this change -> AI Mode will now open publisher links side by side with AI Mode (versus taking you directly to the publisher website). So both AI Mode and the...
Operators Favor Legacy PMSs Over Risky AI Startups
Appfolio and the other "legacy" PMSs are racing to build AI agents before the new AI-native startups can build their own PMSs. For the record, all the operators I know (from people running 100 units to people running thousands) are rooting...
What Hotels Are Getting Wrong About AI
Richard Valtr, founder of cloud‑based PMS Mews, says most hotels are treating AI as a checkbox exercise, adding chatbots and voice assistants without a cohesive strategy. He argues the industry focuses on yield management—price optimization—while neglecting broader strategy management that...
Canada Opens Applications to Build a Public AI Supercomputer
The Canadian government has launched the AI Sovereign Compute Infrastructure Program (SCIP), allocating roughly $890 million CAD (about $660 million USD) to build a publicly owned, large‑scale AI supercomputer. Applications are open to Canadian non‑profits, post‑secondary institutions, or consortia, with a deadline...
Bridge the Visual Gap: How to Integrate Applitools Autonomous with PractiTest
Applitools Autonomous’s AI visual validation can now be tightly coupled with PractiTest through a serverless bridge built on AWS Lambda and DynamoDB. The integration eliminates data fragmentation, manual status updates, and latency by automatically syncing visual test results and links...
Cohere’s Aidan Gomez Chooses Sovereign AI Over Easy Wins
"We're not looking for the easiest path, we're looking for the right path." NEW: Aidan Gomez is trying to build a different kind of AI company at $6.8B-valued Cohere. On The Upstarts Podcast, Gomez talks about his journey from small-town Ontario to...
OpenAI Launches Safety Fellowship to Fund External AI Research
OpenAI announced a six‑month Safety Fellowship launching in September 2026, aimed at external researchers studying AI risks. The program will fund fellows with stipends, compute access, and technical support to explore robustness, privacy, agentic oversight, and misuse prevention. By extending...
Israel’s Unit 8200 Is an Early Adopter of AI in Warfare
Unit 8200, Israel’s largest intelligence formation with about 5,000 soldiers, handles signals, open‑source, cryptanalysis and cyber‑warfare, and is often likened to the U.S. NSA. Historically it contributed to the Stuxnet virus and the 2024 Hezbollah pager attacks. Since the October 2023...
AI Can Improve Scholarly Writing — If We Use It Right
The article argues that AI, exemplified by Claude, can enhance scholarly writing when used as a thinking partner rather than a substitute for intellect. It highlights how senior researchers leverage AI to sharpen arguments, spot patterns, and iterate drafts, while...
What We Learned From a Failed Nota News Experiment
Last fall, Nota launched an experiment with 11 hyperlocal news sites, using AI‑assisted tools to turn public documents into local stories. Contractors deviated from the approved source list, copying material from other outlets without attribution, a breach the company attributes...
Automate Your Job
The post details how a Portland logistics broker reclaimed 15 hours a week by linking OpenClaw, an open‑source AI agent, with OpenAI’s new Codex desktop app. OpenClaw reads vendor PDFs, extracts tracking data, and hands it to Codex, which updates...
Dig Launches Ask-Dig, a Video-Centric AI Social Search Platform
dig introduced ask-dig, a video‑first AI platform that turns social media into a real‑time answer engine. Users type natural‑language questions and the system scans thousands of social posts, extracting narratives, sentiment and emerging signals. Each response is linked back to...
Moloco Launches AI-Powered Performance CTV for App Marketers
Moloco has introduced Moloco Ads for Performance CTV, extending its AI‑driven advertising technology to connected‑TV inventory. The platform optimizes each impression in real time, targeting specific marketing outcomes while leveraging advertisers' chosen mobile measurement partners for attribution. It promises premium,...
3 Mistakes Superintendents Say They've Made in Rolling Out AI
Superintendents from Rush‑Henrietta, San Diego Unified, and Broward County shared lessons at the ASU+GSV conference about missteps in K‑12 AI rollouts. They highlighted three common errors: assuming every leader was prepared, neglecting labor unions and parents, and spending excessive time...
From Static Snapshots to Dynamic Protein Modeling
AlphaFold solved protein structure prediction. That's a snapshot. But biology isn't static. Proteins flex, shift, and interact across time. Drug targets have multiple conformations. Cells don't act alone. The next hard problem is capturing that motion, and building models that reflect...
AI at the Core: Trustworthy Intelligence for an Uncertain Hospitality World
Hospitality executives are moving AI from experimental projects to core commercial functions, yet trust remains a major hurdle. IDeaS research shows nearly 80% of hotel groups use AI and 90% plan deeper investment, but only about 1% rely on it...
A New Way to Explore the Web with AI Mode in Chrome
Google has upgraded AI Mode in Chrome, allowing links to open side‑by‑side with the AI chat, so users can browse and ask follow‑up questions without switching tabs. The update adds a plus‑menu that lets users pull in multiple open tabs,...
Google's AI Mode Update Tries to Kill Tab Hopping in Chrome
Google rolled out an update to Chrome’s AI Mode that embeds the chatbot as a persistent sidebar, allowing links clicked within AI responses to open in the same tab. The change eliminates the traditional tab‑hopping experience, letting users continue the...
OpenAI’s Big Codex Update Is a Direct Shot at Claude Code
OpenAI announced a major upgrade to its Codex agent, enabling it to operate macOS desktop applications autonomously and run multiple agents in parallel. The update adds image generation via gpt‑image‑1.5, new plug‑ins for GitLab, Atlassian Rovo and Microsoft Suite, and...
Google Now Lets You Explore the Web Side-by-Side with AI Mode
Google rolled out an upgrade to AI Mode in Chrome that lets users open linked webpages side‑by‑side with the conversational search pane. The feature works on desktop and mobile, preserving search context while users browse. Google also added the ability...
Opus 4.7: Smarter, More Precise AI Upgrade
Opus 4.7 feels more intelligent, agentic, and precise than 4.6. It took a few days for me to learn how to work with it effectively, to fully take advantage of its new capabilities. Will post a few more tips throughout the...
Why Not All AI “Context” Is Equal
Enterprise AI has reached a turning point as leaders realize that better models alone won’t drive value—context does. While fine‑tuning promises domain alignment, it often fails to capture evolving codebases, security policies, and workflows. Retrieval‑Augmented Generation (RAG) emerged as a...
Accountability in Automated Decisions: The Next Frontier of Tech Law
Enterprises deploying AI‑driven decision systems now face concrete accountability mandates under the EU AI Act and GDPR. Regulators require documented lifecycles, human oversight, traceable data, and contestability mechanisms, shifting responsibility from model explainability to system architecture. The article outlines a...
Millennials, Gen Zers Warm up to AI Shopping Tools
A new Quad and Harris Poll survey shows 62% of Gen Z and Millennial shoppers prefer AI‑powered shopping tools, with 60% of Millennials trusting AI advice more than store associates. However, 75% say they would lose trust if AI results were...
ChatGPT Often Retrieves But Rarely Cites Reddit Pages, Data Shows via @Sejournal, @MattGSouthern
Ahrefs examined 1.4 million ChatGPT 5.2 prompts and found that while roughly half of retrieved pages are eventually cited, pages from a specific Reddit source are cited only 1.93% of the time. The Reddit source accounts for 67.8% of all retrieved‑but‑uncited pages,...
Inside the Bank Where Employees Built 20,000 AI Tools in 18 Months
BBVA rolled out a bottom‑up generative‑AI platform that enabled 120,000 employees across 25 countries to build roughly 20,000 custom GPTs in just 18 months. By allocating only 3,000 licences and empowering a peer‑led "Champions and Wizards" network, daily active usage...
How Tencent Is Using AI to Make Fighting Games More Fluid | GDC Interview
Tencent’s MoreFun Studios is deploying generative AI to fine‑tune character movements in real time for its upcoming fighting title, The Hidden Ones. At GDC’s Festival of Gaming, head of AI Elvis Liu demonstrated how the system can produce fluid animations...
Gemini Adds Personal AI Photo Generation with Google Photos
Another step towards Jarvis -> Gemini is now bringing together Personal Intelligence, Google Photos, and Nano Banana 2 "If you have manually connected Google Photos to Personal Intelligence, Gemini will “use actual images of you and your loved ones to...
Man Used AI to Make False Statements to Shut Down London Nightclub, Police Say
A London businessman pleaded guilty to submitting AI‑generated letters that falsely opposed the reopening of Heaven, an LGBTQ nightclub, after a prior licence suspension. The letters, sent via an encrypted email address, were identified as fabricated by AI‑detection tools and...
The Pulse: ‘Tokenmaxxing’ as a Weird New Trend
Developers at firms like Meta, Microsoft and Salesforce are deliberately inflating AI usage by burning tokens to hit internal metrics, a practice dubbed "tokenmaxxing." At the same time, companies such as Uber have exhausted multi‑year AI token budgets in months,...
The Tightrope Walkers: Conversational AI Must Bridge Modern AI And Contact Center Reality
More than 650 conversational AI vendors compete to serve the roughly 15 million contact‑center agents worldwide. Forrester’s Q2 2026 Wave highlights 14 platforms that successfully combine large‑language models, generative AI, and agentic frameworks while meeting enterprise security, observability, and integration requirements. These...
Thursday April 16, 2026 — Field Note
AI startup Luminai announced a $38 million Series B round to expand its intelligence platform across health‑system operations. The funding, led by PHTI, backs a partnership with the Cleveland Clinic, which serves roughly 15 million patients annually, focusing initially on referral‑management workflows. Luminai...
OpenAI Chooses CrowdStrike for Defender‑Focused AI Security
The top AI labs are building for defenders now. Today @OpenAI selected CrowdStrike for their Trusted Access for Cyber program. CrowdStrike continues to lead the market in secure AI adoption, trusted by AI leaders and organizations of all sizes to accelerate...
AI Is Not Alien, It's Anti‑Intelligence Opposing Human Thought
Hariri is close, but I think he misses the point by calling AI alien. It's better to think of AI as anti-intelligence--intelligence that is antithetical to human cognition. https://t.co/m1ErdtsBgs
XAI Renting GPUs to Cursor – Further GPU Rents Enables Breakeven
Elon Musk’s xAI is set to lease tens of thousands of GPUs to AI‑coding startup Cursor, with hourly rates between $2.5 and $18. Renting 10,000 GPUs would cost $15‑$40 million per month, while 50,000 GPUs would run $75‑$200 million. If xAI rents...
AI Learns Scientific Laws, Not Just Memorized Answers
With Sim2Reason, AI moves closer to how science itself progresses: not by memorizing answers, but by discovering the laws that generate them. 🚀🧠 Read more about the team’s work: https://t.co/CSFqdns1er
Gemini Captures 25% AI Traffic, Validating Google Strategy
🍿 Gemini has now +25% of AI platforms traffic 👇 Google strategy is working well
Sage Haven Launches AI-Moderated Chat App for Kids
Sage Haven, a public‑benefit startup founded by sisters Kate Doerksen and Anne Pizzuti, has launched an AI‑moderated messaging and voice‑calling app for children after a beta phase and a $3 million pre‑seed round. The app assigns each child a dedicated phone...
First Look at My New AI Smart Glasses
I finally got my AI Smart glasses by @Vader_AI_ Check out the unboxing 👀 https://t.co/ldUq81Qk9r
Generative AI Matures Into Six‑Layer Autonomous Agent Stack
The Generative AI ecosystem is evolving into a full tech stack — powering autonomous AI agents. From infrastructure and LLMs to RAG pipelines, agent behaviors and orchestration layers, this framework shows the 6 layers driving next-gen AI systems. Credit: @goyalshalini #AI #GenerativeAI #AgenticAI...
AI Trains Electron Microscope to Match Human Vision
A decade of imaging. Compressed into three months. Here’s how MIT CSAIL & Harvard taught an electron microscope to see like you do: https://t.co/qC3zVck0HF https://t.co/kMTptacRMD
Free AI Prompt Turns Your Data Into Business Lessons
Want my Full Business Reflection Prompt (for free)? This AI prompt pulls from your calendar, Slack, and memory to build a “What Happened” list and then extracts the lessons hiding inside it. Comment the word RECAP below, and I’ll send it over.
Future Is Here: Localizing Products From a Self-Driving Car
It's easy to forget that you're living in the future. But every now and then you see something like this... Responding to a client's complaint by making a major product change (localization) from your phone, talking to your software agent...
AI Risks Undermining Civic Engagement Skills in Higher Education
AI threatens to erode another key aspect of higher ed — teaching students the habits of civic engagement that make democracy work. As students use ChatGPT for discussion posts and routinely consult bots, will they lose ability to build...
China Doubles AI‑Science Compute Without US Chips
China doubles ‘AI for science’ computing scale in 2 months using no US chips | South China Morning Post https://t.co/3H0S5aW1UM
Microsoft’s New Image Model Accelerates Shift From OpenAI
.@Microsoft’s MAI-Image-2-Efficient model accelerates company’s move away from @OpenAI https://t.co/7i04tRYv4p @SiliconANGLE @Mike_Wheatley “Image generation is a forte of LLMs and popular for business and consumers alike. Microsoft’s own AI...” #AIWars #NextGenApps
ADK Skills Streamline Python Agent Development
I should have used our ADK skills last weekend when I built some Python agents. Next time. This post explains what these dev skills offer, and why it makes it simpler for agent builders. https://t.co/qtGb3NZetk https://t.co/tIz6tZccLP
Only 28% of AI I&O Projects Meet ROI, 20% Fail
"Results of a Gartner study released Tuesday reveal that only 28% of AI use cases in infrastructure and operations (I&O) fully succeed and meet ROI expectations, and a full 20% end up failing outright." https://t.co/zFS0Cy2PEy