Today's AI Pulse
Trump flags Anthropic as a national‑security threat before easing stance
President Trump told The Axios Show that Anthropic was briefly deemed a national security threat, prompting the Commerce Department to impose export controls and the Pentagon to label the firm a supply‑chain risk. The administration also barred foreign nationals from the company, actions typically reserved for foreign adversaries, but the tension eased after a recent G7 meeting.
Also developing:
By the numbers: NewGenIVF raises $10M pre‑Series A for K25.ai
Q&A: Box CEO Embraces Shift to ‘Headless’ Software in the Agentic AI Era
Box CEO Aaron Levie says AI agents will become the primary consumers of SaaS, forcing vendors to adopt headless, API‑first platforms. Box, built on an API business model, is expanding its content management system to be fully accessible to agents while retaining seat‑based pricing for human users. Levie notes that security, governance and data control remain essential as agents interact with enterprise content. He likens the AI shift to cloud adoption but far broader, affecting every employee’s workflow.
Most Wildlife AI Focuses on the Ground. This Model Looks up in the Trees
Scientists introduced TropiCam‑AI, an artificial‑intelligence classifier designed to detect tree‑dwelling mammals and birds in neotropical forests. The model can recognize 84 taxa—including 63 species—with an overall accuracy of 95%, and it flags ambiguous images by returning higher‑level taxonomic guesses. Training...
How to Use YouTube’s New AI Update to Take Total Control of Your Video Recommendations
YouTube has launched a "custom feeds" feature that leverages Google’s Gemini AI to let users craft personalized video streams via natural‑language prompts. The tool, initially tested last year, is now available to all U.S. mobile users in English and appears...
ITWebTV Biz: Prosperity with AI Is a Full-Stack Strategy
Huawei Enterprise South Africa’s new CEO Kui Zheng highlighted the continent’s accelerating AI adoption, noting that over two‑thirds of local enterprises are experimenting with generative AI and 15 % have approved tools. He emphasized that AI’s promise hinges on robust data,...
Google Is Not Just Updating Gemini, It Is Building an AI Operating Layer
Google unveiled Gemini 3.5 Flash, Spark, and Omni at I/O 2026, turning Gemini into an AI operating layer that spans Search, Workspace, Cloud, and consumer devices. Flash is a fast, multimodal agentic model with up to one‑million‑token context and 64,000‑token...
SpaceX's C AI Stack Promises 10× Speed over JAX
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible. The potential speed improvement...
Talkdesk Introduces Proactive AI Agents to Drive Growth in Retail and Financial Services
Talkdesk launched industry‑specific proactive AI agents for retail and financial services, extending its Customer Experience Automation platform. The agents automate outbound outreach—recovering abandoned carts, handling product recalls, and accelerating loan, deposit, and collection processes. Built with templatized multi‑agent workflows, they...
A Guide to the New, Wide World of Agentic Advertising and Commerce Protocols
AI agents are reshaping the web, driving a 23.5% YoY surge in automated traffic and a 400% jump in retail‑focused AI visits in Q1 2026. The Model Context Protocol (MCP), now governed by the Linux Foundation, provides the universal adapter that...
Lithuania’s Kopa.ai Raises €2 Million to Scale Its Agentic AI Co-Pilot for E-Commerce Teams
Lithuanian AI startup Kopa.ai announced a €2 million (≈$2.2 million) seed round led by XTX Ventures and Practica Capital. The Vilnius‑based platform, which bills itself as an operating system for e‑commerce teams, already serves more than 2,000 merchants and posted €2 million in...
Warwickshire County Council Seeks AI Partner
Warwickshire County Council has opened a two‑year procurement for AI and automation tools, allocating an estimated £2 million (≈$2.5 million) for the project. The initial spend of up to £1 million (≈$1.25 million) will fund speech transcription, document generation, translation and chatbot services accessed...
AXIOMATIC INTELLIGENCE: NOT YOUR ORDINARY AI
U.S. Army War College researchers warn that data overload and unpredictable large language models are hampering battlefield decision‑making. They propose “axiomatic AI,” a deterministic system that pulls verified facts from trusted Army data sources using formal logic instead of probabilistic...
Investigating the Potential Use of Frontier AI Models for Offensive Cyberattacks: A Human Uplift Study
A RAND‑conducted human uplift study commissioned by the UK AI Security Institute examined whether frontier AI models boost offensive cyber capabilities among lower‑skilled participants. One hundred fifty‑seven volunteers tackled network, OS and vulnerability‑exploitation challenges, half with access to models such...
Google Cloud Introduces AI-Driven Threat Defense for Enterprise Security
Google Cloud unveiled AI Threat Defense, an AI‑native security platform that monitors enterprise AI environments in real time. The service fuses Google Threat Intelligence, Security Operations, and Wiz’s cloud‑AI tools to detect and block AI‑driven attacks such as prompt‑injection, model...
Enterprise AI Deployment Needs Tenfold More Resources than Expected
Take whatever number of people you thought might be in jobs related to AI deployment in the enterprise and multiply it by 10. Then probably 10 again. A major topic that keeps coming up in talking to CIOs across enterprises...
Self-Evolving Agent Skills: SkillOpt
SkillOpt, a new framework from Microsoft and leading Chinese universities, treats agent instructions as trainable external state, applying deep‑learning‑style optimization to text. By imposing a bounded edit budget—called a textual learning rate—the system performs controlled “textual gradient descent,” preventing erratic...
So You Still Think You Can Google?
Google unveiled its biggest search UI overhaul in 25 years, adding an AI‑driven Overview that delivers concise answers and pre‑selects source material. The new layer still shows source links, but it decides which sites count, limiting the traditional list of...
Biased AI Writing Assistants Shift Users’ Attitudes on Societal Issues; Synthetic Sources?
Recent large‑scale experiments show AI writing assistants that supply biased autocomplete suggestions can nudge users’ opinions on societal topics toward the model’s stance. In a sample of 2,582 participants, post‑task surveys revealed measurable attitude convergence, yet most users did not...
Nvidia Posts $81.6 B Q1 2027 Revenue, Cites $200 B AI Chip Market Opportunity
Nvidia announced fiscal Q1 2027 revenue of $81.6 billion, an 85% year‑over‑year jump, and lifted its full‑year outlook. CEO Jensen Huang said the newly unveiled Vera CPU unlocks a $200 billion agentic‑AI market, while Australian data‑center operator IREN signed a $1.6 billion deal for...
Teen Builds AI‑powered Voter Guide, Draws 70,000 Users in Tamil Nadu
Kapil Bhaskar, a 16‑year‑old student, used Claude AI and Emergent to create Nee Yosi, a free bilingual platform that distilled candidate affidavits for Tamil Nadu voters. Built in under a week for roughly ₹1,500, the site logged more than 70,000...
Bear Electric Teams with Baidu AI Cloud to Launch AI‑Powered Smart‑Living Ecosystem
Bear Electric Appliances announced a strategic partnership with Baidu AI Cloud to embed large‑scale AI models, voice interaction and intelligent control into its appliance portfolio. The collaboration aims to turn everyday devices such as air fryers, rice cookers and baby...
Why World Models Must Do More than Simulate: Pony.ai CTO
Pony.ai CTO Lou Tiancheng argues that autonomous‑driving world models must evolve beyond photorealistic simulation. A useful model should act as a training system that captures objectives, dynamics, and interaction with other agents. It must be diagnosable, identifying whether failures stem...
Edgecore (an Accton Company) and Partners Unite at Computex 2026: Powering the All-Photonics AI Era Together
Edgecore (an Accton Company) announced the Edgecore Open Fabric, an all‑photonics AI infrastructure platform built to NTT’s IOWN specifications. The solution combines LCoS‑based optical switching, open‑source networking, composable compute with CXL 3.1, and high‑density liquid cooling capable of 200 kW per rack....
SK Hynix Thermal Solution Helps Boost AI Performance
SK Hynix unveiled a next‑generation thermal solution that leverages graphene‑based heat spreaders to improve heat dissipation in high‑bandwidth memory (HBM) modules used for AI accelerators. Independent testing showed up to a 12% performance uplift in inference workloads and a 15 °C...
Neural Weight Norm Mirrors Kolmogorov Complexity
He finds that neural networks may be biased toward the simplest fit that explains the data OR the neural network with the smallest possible weight norm (that fits the data) must encode the shortest program (that fits the data). "Neural Weight...
AI Narrative Fingerprints Persist Despite Stylistic Prompts
There is a lot being written about the stylistic tells of AI writing (em-dashes, etc.) but this paper looks at AI narrative tells Fascinating differences between AI & human narrative, and asking AI to write in different styles doesn't do much...
AI Applied: Understanding AI's ROI
In this episode of AI Applied, host Jaden Schaefer talks with Connor Grennan, CEO of AI Mindset, about how to measure the return on investment (ROI) of AI initiatives. They argue that ROI should be viewed like any employee salary—by...
Google CEO Sundar Pichai Admits an AI Overview Result Was “More Opinionated than It Should Be” On a Chromebook Search
Google CEO Sundar Pichai acknowledged that an AI Overview result for a “best Chromebook” query was more opinionated than intended. The AI highlighted the Acer Chromebook Plus Spin 714, while the top organic Reddit result suggested a different device. Pichai...
Turning AI Disruption Into Recurring Revenue for MSPs
Pax8 used its Mission Brief Sydney conference to unveil a growth blueprint that helps managed service providers (MSPs) turn AI disruption into recurring, high‑margin revenue. The cloud marketplace now curates 140 vetted vendors—adding 26 in the past year—and supports 4,500...
Machine‑Learning Model Boosts Depression Remission to 55% in Pilot Study
A UC San Diego pilot employing machine‑learning‑guided digital phenotyping reported that 55% of adults with mild‑to‑moderate depression no longer met PHQ‑9 criteria after six weeks, while anxiety scores fell 36%. The findings, highlighted by the Substance Use Disorder Foundation, underscore...
Illinois Lawmakers Just Passed America’s Strongest AI Safety Bill
Illinois lawmakers approved a bill that obliges frontier artificial‑intelligence labs—including OpenAI, Anthropic and Google DeepMind—to undergo third‑party safety audits. The legislation also mandates annual risk‑assessment reports to the state attorney general and imposes fines of up to $10 million per violation....
TeraWulf Stock Jumps 10% After Acquiring Kentucky Data‑Center Campus
TeraWulf (WULF) surged more than 10% after revealing it bought the Muskie Data Campus in Kentucky, a 285‑acre site designed for hyperscale AI and high‑performance computing. The acquisition marks a decisive pivot from pure‑play Bitcoin mining to large‑scale, power‑advantaged data‑center...
Media Research Center Claims AI Chatbots Show Left‑Wing Bias, Raising Trust Concerns
The Media Research Center released a study alleging that major AI chatbots display a left‑wing bias, prompting criticism from right‑leaning observers and fueling broader worries about the credibility of digital news sources. The claim comes amid heightened scrutiny of AI‑generated...
AIC Teams with NVIDIA and VAST Data to Unveil AI Infrastructure at Computex 2026
AIC announced it will display its latest AI storage and compute platforms at Computex 2026, partnering with NVIDIA and VAST Data. The company will also host a strategic panel on June 2 to discuss breaking the memory wall for enterprise...
Dell and Palantir Launch On‑Prem AI Operating System for Secure Enterprise Analytics
Dell Technologies and Palantir announced a joint on‑premises AI operating system at Dell Technologies World 2026, combining Palantir Foundry, Ontology and Dell AI Factory with NVIDIA hardware. The solution promises secure, governed AI workflows for regulated industries that cannot rely...
Alipay Launches AI Wallet and Token Pay to Power Agentic Commerce in China
Alipay rolled out the world’s first AI Wallet and Token Pay on May 26, creating an AI‑centric payments layer for merchants, AI agents and consumers. The launch includes a trust protocol and security system designed to enable “agentic commerce” across...
SpaceX's xAI Pursues New AI‑Compute Clients After $45 B Anthropic Deal
SpaceX's xAI subsidiary is courting additional AI‑compute customers after locking in a $1.25 billion‑per‑month, three‑year contract with Anthropic for its Colossus I data center. The move comes as the company prepares a June 12 IPO and plans to expand its terrestrial...
AI‑Driven Exploits Overtake Password Theft as Top 2026 Breach Vector, Mobile Phishing Surges
Verizon's 2026 Data Breach Investigations Report finds AI‑powered vulnerability exploitation now drives 31% of corporate breaches, eclipsing stolen passwords at 13%. Mobile phishing attacks achieve 40% higher click rates than email, making phones the fastest‑growing entry point.
Rewiring Software Delivery for the Agentic Era
Leading firms are redesigning software delivery around a 24‑hour sprint that pairs human judgment with AI agents executing overnight. By standardizing requirements, architecture, and tooling, agents can generate, test, and package code continuously, delivering three‑ to five‑fold productivity gains while...
Claude Opus 4.8 on GitLab: Complex Agentic Work, Less Disruption
Anthropic’s Claude Opus 4.8 has been added to GitLab’s Duo Agent Platform, targeting complex, multi‑step agentic workflows. The model promises tighter execution, fewer manual corrections, and stronger performance on coding, document drafting, and data analysis. It also introduces mid‑conversation system...
Move First or Fall Behind: How AI Is Rewriting the Rules of Banking
McKinsey partners Ido Segev and Roberta Fusaro warn banks that AI is no longer a pilot but a strategic imperative. Rapid LLM upgrades, growing customer willingness to delegate tasks, and a push toward 20‑30 virtual agents per human employee are...
AI-Driven Nanotweezers Bring Milk Vesicle Analysis Into Sharper Focus
Researchers unveiled an electrohydrodynamic nanotweezer platform that combines AI‑driven image analysis with label‑free interferometric scattering microscopy to trap, size and sort milk‑derived extracellular vesicles (EVs). The system immobilizes thousands of vesicles in seconds, extracts diffusion‑based size estimates, and infers refractive...
Faultformer: A CNN–Transformer Multi-Domain Feature Fusion Network for Intelligent Bearing Fault Diagnosis
FaultFormer, a hybrid CNN‑Transformer network, fuses statistical, frequency‑domain and deep features to diagnose rolling‑bearing faults from vibration data. Tested on the CWRU bearing dataset across ten fault types and load conditions, it achieved a mean classification accuracy of 99.98 % over...
Gemini Managed Agents Blend Hosted Environment with Tool‑Use Loop
"The useful part of Gemini Managed Agents is the combination of a hosted environment and a built-in tool-use loop."
Safe AI Agents Need Strict Permission Boundaries
I watched Claude run rm -rf / to "test permissions" and report back cheerily that it had failed. I went from WTF to absolutely-fucking-not in about a second. Good fences make good agents. Wrote up sandvault + the /sv skill I...
Anthropic’s Hidden Strategy Could Reshape the Entire AI Market #164b
Anthropic quietly upgraded its Opus tokenizer, turning the unchanged $5‑$25 per‑million‑token rate into a 30‑50% effective price hike for users. At the same time, the company is deepening its enterprise AI infrastructure, making migration costly for large‑scale customers. Open‑source alternatives...
Global Trade Linkages and the Cross-Country Distribution of the Gains From AI
A new OECD‑backed study models how artificial‑intelligence (AI) adoption reshapes global welfare through trade. It finds AI use in core business functions averages 4 % across OECD economies, with a ten‑year horizon ranging from 0.2 % in Mexico to 7 % in Ireland....
Multimodal Emotion Recognition From Images and Text Using Deep Learning and Explainable AI
Researchers introduced ITMER, a unified framework that combines textual and visual data for emotion classification. The system evaluates three configurations: text‑only transformers, image‑only CNNs, and a fused image‑text model. In benchmark tests, the multimodal approach achieved up to 92 % accuracy...
Regula Warns Most Organizations Don’t Know What Bots Are Doing, or Why
Regula’s "New Shape of Identity Threats" report warns that most organizations cannot reliably identify AI‑driven bots or synthetic actors infiltrating their systems. These agents mimic legitimate user behavior, rendering traditional human‑centric verification tools ineffective. The study finds that one‑quarter of...
Cyber Defense in the Era of Frontier AI: Insights From Mythos and GPT 5.5 Cyber
Zscaler evaluated frontier AI models Anthropic Mythos and OpenAI GPT 5.5 Cyber, finding they can reason across attack paths and generate multi‑step exploit workflows. Using three test harnesses—black‑box, artifact/code, and gray/white‑box—the models surfaced twice as many high‑severity findings and did so in...
Agents Now Configure Themselves via CLI, Hands‑free
We are officially entering the "agents configuring agents" era. Instead of wrestling with local MCP settings, you can now hand the shell directly to Claude or Cursor, run a single CLI command, and watch it onboard itself hands-free. 🦾 #agents...