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
Rebellions Raises 640 Billion Won in Pre-IPO Round as First Korea National Growth Fund Pick
South Korean AI chip designer Rebellions Inc. closed a 640 billion won pre‑IPO funding round, valuing the company at 3.4 trillion won. The round, led by the state‑backed Korea National Growth Fund with 250 billion won, also included contributions from KDB and Mirae Asset. Proceeds will fund mass production of the Rebel100 chip, a chiplet‑based processor supporting HBM3E aimed at AI data‑center workloads. The company is already testing the chip with global hyperscalers such as Meta.
AI Makes Music Creation Instant, Challenges Originality
AI is lowering the barrier to making music to almost zero. Tools like Suno let anyone generate songs from simple prompts, turning creation into something instant and accessible. That opens the door to massive creativity, but also raises questions about originality...
The Risks of AI Recording Devices and Note-Taking Assistants in the Classroom
U.S. universities are witnessing a surge in AI‑driven note‑taking apps and smart‑glass recorders, tools that capture audio, video, and biometric data without clear consent. Platforms such as Otter.ai operate with opaque data‑retention policies, while devices like Meta’s smart glasses embed...
Oracle Reportedly Lays Off Thousands of Employees to Bankroll Its Massive AI Infrastructure Bet
Oracle announced a massive workforce reduction, targeting 20,000 to 30,000 positions, to free up roughly $10 billion in cash flow for its AI infrastructure push. The cuts follow a $50 billion capital raise plan that has left the company in debt and...
Brain Corp Updates Floor-Cleaning Robots with Adaptive AI that Removes Route Training
Brain Corp unveiled BrainOS Clean 2.0, integrating SelfPath AI into Tennant’s X‑series floor‑cleaning robots. The update lets machines autonomously generate and adjust routes, removing the need for manual training. Early deployments show 22% higher coverage, 55% greater autonomy, and deployment speeds more than...
The ROI of Intelligence: How AI Agents for Business Operations Are Delivering Value
G2’s review of over 2,700 AI‑agent implementations shows rapid adoption, with 88% of users praising productivity gains and an average ROI realized in just over five months. Automation, conversational intelligence, and real‑time analytics emerge as the most valued capabilities, while...
New Work in the Age of AI Doesn’t Need a Government Planner
A new NBER working paper by David Autor and co‑authors shows that one in five American workers now hold jobs that did not exist before 1970, and these "new work" roles command a wage premium, especially in technology‑linked occupations. The...
Vertiv Expands Ohio Manufacturing to Support AI Data Center Cooling Demand
Vertiv announced a roughly $50 million investment to expand its Ironton and Westerville, Ohio facilities, boosting manufacturing capacity for advanced liquid‑cooling and chilled‑water systems. The Ironton plant’s output is slated to increase about 45 % by Q2 2027, enabling faster response to AI‑driven,...
Leaked Claude Code to Be Reverse‑engineered and Open‑sourced
Leaked Claude Code will be dissected, analyzed, and reverse-engineered. Every pattern and trick will end up in open-source projects. Chinese model makers will absorb whatever they can.
AI News Roundup – OpenAI Discontinues Sora App and Shifts Priorities, Arm to Develop AI CPUs, Study Warns Against Sycophantic...
OpenAI announced it will discontinue its Sora video‑generation app, citing high compute costs and limited monetization, as it pivots toward business‑focused offerings ahead of a planned 2026 IPO. Arm Holdings revealed it will produce its own AI‑optimized CPUs, the Arm...
C‑Level Commitment Needed to Bridge AI Adoption Gap
If you find yourself like so many organizations struggling with the growing gap between AI's potential and your actual, daily leverage of that potential, Gary Survis says you likely have one to three problems: 1️⃣ Your people are not business...
CFOs Unlock Expansion via Clean Data and AI
I really suck at software demos, but hopefully that didn't diminish my new Revenue Intelligence feature I demo'd today. I may be able to methodically explain a P&L, but I realized today that I need training on software demos. Can...
Claude Code Users Say They’re Hitting Usage Limits Faster than Normal
Anthropic has confirmed that Claude Code users are exhausting their usage limits far faster than before, with some prompts consuming up to 10% of a monthly quota and simple greetings using 2% of a session. Users on the $100‑per‑month plan...
She Was Working Until 3 A.m. Every Quarter — What She Built Next Should Be a Lesson for Every CEO
A contract‑operations specialist at a large enterprise built an AI agent, "Connie," to automate data‑pulling, document processing, and workflow routing that previously kept her working past midnight each quarter. The tool accelerated her contract processing tenfold, improved output quality, and...
Anthropic Accidentally Publishes Claude Code Source Code for Anyone to Find
Anthropic unintentionally released over 500,000 lines of Claude Code source on the public NPM registry, exposing more than 1,000 internal files. The leak, attributed to human error rather than a security flaw, included details on unreleased models and features. No...
RSAC 2026: Reflections on a Security Show That Became an AI Showcase
RSA 2026 turned into an AI‑centric event, with security serving as the backdrop. Attendees heard concrete announcements on runtime guardrails, red‑team testing, and lifecycle controls, signaling that AI security is moving from theory to operation. Vendors touted unified integrations of...
How to Build and Evolve a Custom OpenAI Agent with A-Evolve Using Benchmarks, Skills, Memory, and Workspace Mutations
The tutorial walks through building a custom OpenAI‑powered agent using the open‑source A‑Evolve framework in Google Colab. It shows how to set up the repository, define a strict system prompt, create a miniature benchmark of text‑transformation tasks, and implement a mutation...
SigReg Loss Boosts Stable Pixel‑based World Models
Paper review: LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels https://t.co/TpFFnwPWkc Nice clean github: https://t.co/HOuqEf0HaF This is the application of the LeJEPA results to world models, trained offline on experience from three different robotics style tests with one to two million steps in...
Oracle and NetSuite Bring AI Into Restaurant Operations as Industry Seeks Unified Systems
Oracle and NetSuite unveiled an AI‑driven restaurant operations platform that consolidates financial, inventory, procurement and workforce data into a single cloud environment. By integrating point‑of‑sale inputs from Oracle Simphony, the solution offers a real‑time "single source of truth" for operators...
OpenClaw Has 500,000 Instances and No Enterprise Kill Switch
OpenClaw, an AI‑driven personal assistant, has exploded to roughly 500,000 internet‑facing instances, with more than 30,000 showing clear security gaps. A UK CEO’s unencrypted OpenClaw workspace was listed for sale on BreachForums, exposing conversations, production databases, API keys and personal...
AI Agents Turn Environment Into Attack Surface
AI agents don’t just inherit LLM risks — they amplify them. I came across an interesting paper: “AI Agent Traps.” Most discussions still focus on prompt injection. That’s already incomplete. Because once agents have: - autonomy - persistence - tool access...
Schneider Electric Federal Solutions Deliver AI-Ready Data Centers
Schneider Electric Federal is delivering AI‑ready data‑center solutions that address the unprecedented power density required by federal agencies. The company emphasizes turnkey, utility‑to‑the‑rack offerings and public‑private partnerships to modernize outdated infrastructure. A flagship $114 million energy‑savings performance contract at Naval Base...
Google Drive Expands AI Ransomware Detection, File Recovery to More Users
Google has moved its AI‑powered ransomware detection and built‑in file recovery for Drive from beta to general availability. The new model claims to spot 14 times more threats and automatically pauses Drive for desktop syncing when encryption activity is detected. A...
Caltech Researchers Claim Radical Compression of High-Fidelity AI Models
Caltech professor Babak Hassibi and his team at PrismML announced a 1‑bit large language model that compresses model size by orders of magnitude while preserving benchmark performance. The technology, unveiled Tuesday, was released as open‑source code, allowing developers to integrate...
Rocketlane Launches Nitro to Automate Professional Services Delivery
Rocketlane introduced Nitro, an AI‑driven execution layer that embeds autonomous agents into its Professional Services Automation (PSA) platform. The new module automates repeatable delivery tasks such as configuration, data migration, testing, and documentation, linking execution directly to resource tracking and...
April 2026 ITL FOCUS: Underwriting
Generative AI, first introduced to underwriting in late 2022, has rapidly accelerated efficiency by automating data collection and triaging submissions. Recent AI agents now perform actions on behalf of underwriters, enabling continuous underwriting that flags real‑time changes such as a restaurant...
AI Extracts Clean Data From Crumpled Receipts Automatically
Perceptron AI @perceptroninc walked me through something that solves a problem most tools still get wrong. Physical receipts → structured data. Automatically. They ran a demo: photos of crumpled, faded, oddly-formatted receipts. Line items, totals, dates, and vendor extracted cleanly — output...
AI Disruption Fears Hitting Secondaries Portfolios Along With War
AI‑driven disruption and heightened geopolitical risk are prompting secondary market investors to re‑evaluate private‑equity fund portfolios. Sellers face tighter pricing as buyers demand deeper due‑diligence on AI‑exposed assets. War‑related supply chain shocks further compress valuations, especially in consumer‑focused funds. The...
AI Cuts Drug Discovery From Years to Months
I joined Sam Fazeli for the latest episode of Bloomberg Intelligence's Vanguards of Healthcare podcast and talk about our partnership with Eli Lilly and why I believe algorithmic approaches are reshaping the earliest stages of pharmaceutical R&D. The biggest impact...
What Separates a Good Finance Agent From a Weekend Project? Inside Sage's AI Architecture with CTO Aaron Harris
Sage’s CTO Aaron Harris detailed a finance‑focused AI architecture that relies on small, domain‑specific language models rather than generic frontier models. By training 7‑billion‑parameter models and wrapping them in an "arbiter" firewall, Sage reduces inference costs by roughly 97 percent...
Choose Builder or Investor: No Sidelines in 2026
Wealth creation has never been more attainable than in 2026. Right now everyone has two choices: 1) Be the builder - use the tools like Claude code and codex. Find problems, build apps, help with ai migration 2) Be the investor...
AI Identifies Multiple Dementias From One Blood Sample
Researchers at Lund University have unveiled a deep joint‑learning AI model that can simultaneously identify five neurodegenerative conditions—including Alzheimer’s, Parkinson’s, ALS, frontotemporal dementia, and prior stroke—from a single blood sample. Trained on the Global Neurodegenerative Proteomics Consortium’s database of over...
AI Boosts Grades, but Erodes True Learning
🚨 The AI Learning Trap When artificial intelligence helps students succeed but not learn. 👉AI boosted student performance 40 percent, but scores dropped 17 percent once the tool was removed. 👉Effortful learning builds critical brain connections that frictionless AI delivery may bypasses. 👉AI...
DHS RIVR Tests Selfie Verification and Liveness Detection on Consumer Smartphones With Five Vendors Meeting All Goals
The U.S. Department of Homeland Security’s Remote Identity Validation Rally (RIVR) completed a large‑scale test of selfie‑to‑document matching and liveness detection on consumer smartphones. Sixteen vendors were evaluated with 1,632 volunteers using iPhone 14, Google Pixel 7 and Samsung Galaxy...
AI Detects Atomic Defects, Revolutionizing Material Design
AI is now being used to see what was previously invisible. MIT researchers have developed a model that can detect atomic-level defects in materials, helping improve strength, heat transfer and energy efficiency. This is where AI goes beyond data. It starts to...
Tiny AI Model Now Runs Directly on iPhone
Small model runs on your iPhone. Pretty well I hear from one of the people who built it.
Wiliot Builds Its Physical AI Supply Chain Platform on Databricks to Operationalize Item-Level IoT Data
Wiliot announced a Built‑On partnership with Databricks, moving its Physical AI supply‑chain platform onto the Databricks lakehouse. The shift lets the company ingest and govern massive item‑level IoT Pixel data streams in a unified environment. By leveraging Databricks’ compute and...
AI Governance Gaps Enable Rogue Agent Threats
Governance around AI and data is still a significant problem. Many vendors still hesitate to talk about it. What Happens When AI Agents Go Rogue? https://t.co/5TE3g0hXng #CIO #CISO #AI #cybersecurity
Founders Becoming Automated: A Documentary-Worthy Revolution
Built startups with human founders. Built copilots for teams. Now watching founders get automated. What @marik and the @feltsensefund team are pulling off is unlike anything I have seen. There should be a documentary about this.
Bet365 Partners with TestMu AI to Accelerate Global Release Velocity with Agentic AI Quality Engineering
bet365 has partnered with TestMu AI, the first full‑stack Agentic AI Quality Engineering platform, to consolidate its software testing across a single interface. The collaboration gives bet365 access to over 3,000 browser/OS combinations and more than 10,000 real iOS and...
Human Judges Teach AI to Appreciate Aesthetics
Machines do not naturally have taste. By having human experts rate AI on actual aesthetics rather than raw accuracy, the folks at @Contra labs are building something truly special 👀 https://t.co/scY1aEj3T0
AI Safety Cannot Depend on Secrecy and Control
The idea that “AI safety” could be based on secrecy and control has been fatally falsified.
Claude Code's Source Code Leaks Via Npm Source Maps
A security researcher uncovered the entire Claude Code repository after source maps in its npm package exposed a Cloudflare R2 bucket containing every file. The leak reveals a sophisticated architecture: a 40‑tool plugin system, a 46,000‑line query engine, multi‑agent “swarms”, an IDE...
AGI Declaration Inevitable, Yet Current AI Lags Human Intelligence
It's inevitable that the labs will declare AGI at some point. It's also clear that what we have now, while utterly amazing in so many ways, still lacks significant capabilities of human intelligence.
Healthcare Embraces AI Twice Faster Than Other Sectors
Healthcare is adopting AI at 2x the rate of other industries. Lenovo’s Healthcare CTO calls it “assistive intelligence.” What that means for the hospital of the future. 👉 https://t.co/eGplzkBn4T @Lenovo @DukeHealth #CES2026 #HITSM
HIMSSCast: Nurturing National Standards for AI in Patient Care
The Nursing and Artificial Intelligence Innovation Consortium, led by FSU Dean Jing Wang, is tackling the gap between laboratory‑tested AI tools and real‑world nursing workflows. It aims to create national standards that embed nurses in AI‑enhanced clinical processes, ensuring predictions...
AI Regulation Mirrors Nuclear Non‑proliferation’s Flawed Moratorium
Good piece on the problem with a datacenter moratorium and the limits of nuclear non-proliferation as an analogy to regulating AI https://t.co/UVjiO0pBBM
Apple's Siri AI Chatbots Face Significant Risk
Apple might soon add AI chatbots that run inside Siri, but there could be a lot of risk involved. https://t.co/FIZilnbhUd
Kenyan AI Startup Bets on Local Dialects as Proof Gap Remains
Map Maven GMB, a Nairobi‑based AI startup founded in 2025, has launched Kaya, a 70‑billion‑parameter LLaMA‑based large language model fine‑tuned on Kenyan dialects, alongside a voice agent called Sauti deployed at Natcon SACCO. The firm also offers Daraja Connect, a...
AI Agent Identity Layer: Next Cybersecurity Frontier
I recently joined @reckless on @DecoderPod to discuss the “SaaSpocalypse,” the future of software, and why the identity layer for AI agents could become the biggest category in cyber. Really enjoyed this conversation: https://t.co/afZ84f2ymM