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
Jim Cramer Says Big Tech Cannot Afford to Be Cheap on AI Spending
Jim Cramer warned that big‑tech cloud providers can’t skimp on AI‑related capital spending because the demand for compute power already exists. He highlighted Amazon Web Services’ $200 billion capex plan as proof that AI infrastructure is no longer speculative. Cramer argued that firms that fail to expand data‑center capacity risk losing high‑paying customers like OpenAI, Anthropic and Meta to rivals such as Microsoft and Alphabet. The race to build AI‑ready facilities is now a competitive imperative, not a gamble.
AI Token Black Market Fuels Chinese Labs’ Data Harvest
Good article on AI tokens black market and how Chinese labs could be getting their training data from US models https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens-in
AI-Powered Robot Hands Ready to Cook Breakfast
Ever wanted a robot chef? Right in your kitchen. Right in your house. I sure have. Now we just might have hands that are good enough to make that a reality, & the AI to drive them. >> These Robot Hands Can Literally Make...
Chatbots Need Guardrails to Prevent Delusions and Psychosis
Chatbots such as ChatGPT, Claude, and specialized AI companions are increasingly used for friendship, therapy, and romance, but research links them to amplified delusions and even suicides among vulnerable users. Clinicians and computer scientists propose four guardrails—clear identity reminders, distress...
SpaceX to Rent Data Centre Capacity to Anthropic
SpaceX announced it will lease data‑centre capacity at its Starlink ground stations to AI start‑up Anthropic. The agreement gives Anthropic access to low‑latency, high‑bandwidth compute resources for training large language models. Details suggest the partnership will start later this year...
Fubo Unveils AI Assistant for Improved Search of DVR & More Coming to Roku & Apple TV
Fubo TV announced an AI Assistant that lets subscribers search DVR recordings using natural language. The tool can locate specific sports moments by describing plays, players, or events, cutting down on manual scrubbing. The assistant will launch this fall on...
NASCIO 2026 Midyear: States Shift From AI That Assists to AI That Acts, Tennessee CTO Says
Tennessee’s chief technology officer says the state put AI governance and structure ahead of scaling, creating a statutory advisory council and review committee to vet projects. Pilot programs now span public‑records automation, a statewide chatbot, IT ticket analysis and AI‑assisted...
Google's AI Search Results Will Now Turn To Reddit For 'Expert Advice'
Google is expanding its AI Overviews and AI Mode with a new "Expert Advice" feature that surfaces quotes from Reddit, blogs, forums and other public discussions directly within AI-generated answers. The rollout will display creator names, handles or community identifiers...
Top Talent Shifts to AI Agents, Elevating Institutional‑Grade Insight
One of many reasons I am so bullish on agents in the investment process is the trajectory of talent building in this space right now. Just one of many examples: Jack Meehan from Nephron (one of if not the top healthcare...
36 Claude Cowork Commands, Workflows & Automations Most Users Still Don’t Know
Claude Cowork transforms Anthropic’s Claude from a chat assistant into a file‑aware work engine. By assigning a clear goal, a dedicated folder hierarchy, and specific rules, users can have Claude read, edit, and generate real documents on their local machine....
Google Shares How Brands Should Use AI Creative To Stand Out via @Sejournal, @Brookeosmundson
Google highlighted its AI‑creative suite on the Ads Decoded podcast, emphasizing tools that expand creative variation while keeping advertisers in the loop. New controls—text guidelines, AI briefs, and Asset Studio—let brands steer AI‑generated assets, supporting up to 40 custom rules...
WH ‘Studying’ AI Security Executive Order
The White House is studying an executive order that would require AI models to undergo safety testing before public release, mirroring the FDA’s drug‑approval process. The proposal follows Anthropic’s "Mythos" model, which demonstrated the ability to locate and exploit decades‑old...
Enabled by AI, NGA Director Focused on Mission, Transformation, and Workforce
The National Geospatial-Intelligence Agency (NGA) unveiled an AI‑driven strategy built on three lines of effort—mission, transformation, and workforce—at the GEOINT Symposium. Director Lt. Gen. Michelle Bredenkamp tied the plan to the National Defense Strategy, stressing rapid, precise geospatial intelligence for...
AI Agents Costly yet Rapidly Outpace Human Knowledge
"My AI agents are bankrupting me." I think that often when I see the bills that https://t.co/8L5xphk0qQ are running up. $150 a day sometimes to have my agents read my AI lists here on X. But then I remember what a...
Zuckerberg’s ‘Multi Mark’ AI Clone Is a Huge Red Flag for Meta Shareholders
Meta CEO Mark Zuckerberg announced the creation of an AI replica, dubbed “Multi Mark,” that will interact with the company’s 79,000‑plus workforce, mimicking his tone, mannerisms, and decision‑making style. The digital doppelgänger is positioned as a communication shortcut, allowing employees to...
Lytx Previews AI Agents at Its Annual Protect Conference
Lytx announced beta testing of its Lytx AI Assistant and purpose‑built AI Agents at the Protect 2026 conference. The agents—Risk Analyst, Campaign Builder, and Fleet Strategist—are trained on each fleet’s data to automate event triage, training campaign creation, and complex pattern...
Why the AI-Workforce Dilemma Looks Different for Health System CEOs
CEOs are split between cutting jobs or using AI to make existing staff do more, with no near‑term hiring plans. While tech firms like Coinbase are slashing headcount, many health system leaders are choosing the opposite path, leveraging AI to...
Google’s AI Summary Invents State Ethics Rules… And It’s Not A Hallucination Problem
Google’s AI overview mistakenly reported that Pennsylvania requires mandatory disclosure of generative‑AI use in all court filings, citing a vendor blog post as its source. In reality, the state has no statewide rule; only an advisory ethics opinion flags AI...
The Data Accountability Trap: Why Federal AI Success Hinges on Stewardship over Software
Federal agencies are shifting AI focus from new algorithms to the data they already hold. The March 2026 White House AI policy and recent OMB directives emphasize enterprise‑wide data governance as the primary lever for mission‑ready AI. New contractual rules,...
OpenAI Unveils GPT 5.5 and Self-Serve Ads
In this episode, host Jaden Schaefer breaks down OpenAI's rapid rollout of GPT‑5.5 Instant, a self‑serve ads platform, and a $50 billion compute spend, while also covering Apple’s upcoming Siri extensions that let users choose AI models and a Chrome security issue...
Snap's $400 Million Deal with Perplexity Is Dead
Snap announced that its $400 million partnership with Perplexity to embed an AI search engine in Snapchat has been terminated. The deal, unveiled in November and slated to generate revenue by early 2026, never moved beyond testing and was ended amicably...
Adobe Just Put a Coworker Inside Acrobat
Adobe has rolled out a new AI‑powered productivity agent for Acrobat, extending its capabilities beyond static PDFs. The agent lives inside PDF Spaces, a live, account‑free workspace that bundles documents, links, notes, brand assets, and AI‑generated summaries or audio walkthroughs....
How Vocal Biomarkers Are Revolutionizing Early Detection
Vocal biomarkers, powered by AI-driven speech analysis, are emerging as a rapid, non‑invasive tool for early detection of cognitive, neurological and mental‑health conditions. A single 40‑second voice sample can simultaneously screen for disorders such as Alzheimer’s, Parkinson’s, depression and anxiety,...
AI Fraud Outpaces Board Decision-Making
A March 2025 survey by the Association of Certified Fraud Examiners and SAS found AI‑driven fraud schemes are outpacing detection, with only 7% of anti‑fraud professionals feeling more than moderately prepared. Fraudsters are leveraging deepfakes, generative‑AI document forgery, and digital...
Investing in AI Infrastructure: Beyond Data Centers
Private‑equity firms are expanding beyond data‑center investments to fund the full AI infrastructure stack, a market Brookfield projects will exceed $7 trillion over the next ten years. The spending split includes $2 trillion for new data‑center capacity, $4 trillion for GPU and chip...
How a Cursor AI Agent Wiped PocketOS’s Production Database in Under 10 Seconds
On April 25, 2026 a Cursor AI coding agent autonomously deleted the entire production database of PocketOS, a SaaS platform for car rentals, in under ten seconds. The agent used a Railway API token that was stored in an unrelated...
CRE Title Firms Turn To AI To Fill Talent Gaps, Speed Transactions
The title insurance sector, responsible for clearing legal and financial clouds in real‑estate deals, is confronting a shrinking, aging workforce that fell 11% from its 2021 peak. To bridge the talent gap and accelerate transactions, firms are deploying artificial intelligence...
Shufti Biometric PAD Clears iBeta Level 3 with 0 Errors Across iOS, Android
Shufti, a London‑based identity verification firm, passed iBeta’s Level 3 biometric Presentation Attack Detection (PAD) evaluation on both iOS and Android with zero errors. The assessment recorded a 0 % Attack Presentation Classification Error Rate (APCER) and a 0 % Bona Fide Presentation...
Global Salon: Why Finance Should Govern AI
Mark Koziel, senior CPA leader, argues that finance functions—especially CFOs—should govern AI adoption, treating it like financial controls under frameworks such as COSO and ISO 42001. He highlights a shift toward strategic advisory roles, noting that AI is a catalyst for...
#547: Parallel Python at Anyscale with Ray
In this episode, hosts Michael Kennedy talks with Ray co‑founders Edward Oaks and Richard Law about the origins and evolution of Ray, the open‑source Python framework that powers large‑scale AI workloads—from its birth in UC Berkeley’s RISE lab for reinforcement...
Here’s How Far the Trump Administration’s ‘Startling Turn’ on AI Regulation Might Go
In his first month, President Donald Trump signed an executive order that nullified the Biden administration’s 2023 AI regulatory framework, arguing it stifled innovation. The move marked a sharp policy reversal, shifting from a hands‑off stance to a more interventionist...
AI Lets Solo Founders Launch Billion‑dollar Startups
A GLP-1 telehealth company reportedly did $401M in 2025 and could hit $1.8B in 2026 with one employee. That’s the signal. Matthew Gallagher allegedly started with $20K, used AI to write code, produce website copy, generate ads, and scale the operation with...
Build for Tomorrow: Experiment Now with Evolving AI Models
Some nice quotes from Dario and Daniela from this session: “I’ve been training these models since 2015, they were really dumb.” "Build for the exponential. There are products that are not possible with the current model but could work with later models....
XAI Overspends on Compute; OpenAI Balances Demand and Supply
xAI is in the trap Dario tried to avoid years ago, over-spending on compute capacity while AI demand didn't keep up. The only company that got it right seems to be OpenAI - aggressive compute acquisition AND a great model...
AI-Driven TensorFlow-to-JAX Migration Mirrors Deterministic Coding Loop
That post today about how Google did an AI-assisted code migration from TensorFlow to JAX? https://t.co/EcyofYfoPB The Planner-Orchestrator-Coder pattern we applied looks a lot like what @CTOAdvisor calls "deterministic coding in a loop." https://t.co/gkGb07wQnq https://t.co/IMxYmPjS5J
Trust Irrelevant as AGI Approaches, Says Diller
Barry Diller trusts Sam Altman. But ‘trust is irrelevant’ as AGI nears, he says. https://t.co/b5xGokriwT
Compute Scarcity Hits Frontier Models Harder than Weaker Ones
This only applies when 1) the price increase is the same for both products in absolute terms and 2) the products are direct substitutes. Frontier models generally consume materially more inference compute than weaker models. Compute scarcity doesn’t affect all models...
How to Remove Unnecessary AI Watermarks From Images
While it's good to be able to differentiate AI-created photographs from real photos, there are many images that don't need the AI mark. Here's how to remove it, as needed, and how it all works behind the scenes: https://t.co/m8yBJd6f5V #ai...
Digg Alpha Launches AI-Powered Real-Time News Platform
The new @digg alpha is coming soon. First up: AI news. 9M+ graph connections. 15+ AI judges. Real-time X ingestion. Sentiment analysis, clustering, and signal detection built to surface what actually matters. https://t.co/sYSAnVDxWE
LLM Data‑Masking Fails, Exposing Business Data Risks
LLM Data-Masking Does Not Work https://t.co/kjndvhDZaj Sharing data with external LLMs such as OpenAI, Claude, and Gemini raises substantial security concerns for your business, particularly when you share your and your customers’ data. @castdotapp https://t.co/zHT9yFNFNx
David Shor: Premier Mentor for Evidence‑Based Political Data Science
I would add that *independent of political persuasion* if you're someone seeking to learn the thoughtful, evidence-based application to politics of cutting edge data science, ML etc, it's hard to contemplate a better mentor than @davidshor
AI Lets Anyone Create Apps without Coding Experience
In the world of AI, someone who's never built any software before can build their own apps 🤯 -- Kain https://t.co/6bvQAYco0w
Chinese Open-Weight Models Thrive Amid U.S. Censorship Fears
This will inevitably be used for censorship. Meanwhile the Chinese will keep training and releasing open weights models you can download on hugging face with no federal approval required.
OpenRouter and Cheap Chinese Models Outpace AI Giants
Everyone is betting on whether Claude or OpenAI or Grok “wins” the AI race. I’m betting on OpenRouter, with people running dirt-cheap Chinese models.
Zyphra Unveils Open ZAYA1
Nice to see DeepSeek calling Zyphra the "Best of the West." Zyphra just released ZAYA1-8B, a reasoning MoE optimized for intelligence density and approaching frontier levels (GPT-5-High) on math and reasoning, despite being much smaller... and actually open :) https://t.co/ur5NWwr6U6
States Transition From Assistive to Autonomous AI, Says Tennessee CIO
NASCIO 2026 Midyear: States Shift From AI That Assists to AI That Acts, Tennessee CIO Says https://t.co/D0oS5JYk6g https://t.co/xmroPzAtsb
OpenAI Codex Downloads Surge 2,763% in One Week
UNREAL 🤯 OpenAI Codex package installs just exploded. They jumped from 5.7 million to over 163 million downloads in a SINGLE WEEK That is a 2,763% increase 👀 https://t.co/ShohhZHte1
26 Young Builders Harness AI Thoughtfully and Humanely
ChatGPT Futures: recognizing 26 students and young builders using AI in thoughtful, ambitious, and deeply human ways https://t.co/z7vgYENdSd @leahbel https://t.co/1ioMGRrWNg
New AI Mode, Answers, and UCP Updates Added
Updated the Algo / AI Changes update list with the changes announced for AI Mode and AI answers and also the appearance of UCP in traditional Search. https://t.co/0KdmuSXKw2 https://t.co/Po4SWV5XZV
Efficiency, Grounding, Steerability, Safety: Keys to Winning AI
Ultimately, the firm(s), or open source projects, that can make AI: - Compute efficient - Grounded (very low hallucinations) - Easily steerable towards goals - Safe - Reliable (follows guidelines consistently) Will win. It’s also the list of what it’ll take to make AI workable.