Today's AI Pulse
-1.png)
Anthropic shuts down Fable 5 and Mythos 5 models under U.S. export controls
Anthropic abruptly disabled its flagship Fable 5 and Mythos 5 models to comply with a U.S. export‑control directive, underscoring the vulnerability of firms that depend on closed‑source AI services. The move triggered a sharp rally in Chinese open‑source offerings as investors shifted to downloadable models that can run on‑premise.
Also developing:
By the numbers: Respond raises $62.5M Series B
OneStream Expands AI Strategy, Adding Interoperable Agents
OneStream unveiled a finance agentic layer that lets its AI agents interoperate with leading large‑language models such as OpenAI’s ChatGPT, Anthropic’s Claude, and Microsoft’s Copilot. The layer relies on an open model context protocol to provide secure, governed access to OneStream data while preserving role‑based permissions and audit trails. It also introduces a new forecast agent that answers plain‑language questions about financial projections. The announcement aligns OneStream with broader industry moves toward AI interoperability, exemplified by SAP’s recent autonomous‑enterprise push.
AI Scribes: Hype vs Proven Efficiency in Healthcare
I used 𝑠𝑐𝑒𝑛𝑎𝑟𝑖𝑜 𝑎𝑛𝑎𝑙𝑦𝑠𝑖𝑠 to analyze the near future of AI scribes in their promise of replacing administration. Key driver: Growing evidence in studies Major uncertainty: Acceptance by medical professionals 𝐒𝐜𝐞𝐧𝐚𝐫𝐢𝐨 #𝟏. 𝐃𝐚𝐧𝐠𝐞𝐫𝐨𝐮𝐬 𝐇𝐲𝐩𝐞 (𝐇𝐢𝐠𝐡 𝐚𝐜𝐜𝐞𝐩𝐭𝐚𝐧𝐜𝐞 / 𝐋𝐨𝐰 𝐞𝐯𝐢𝐝𝐞𝐧𝐜𝐞) Clinicians adopt AI scribes enthusiastically...
Free AI Rendering Studio Replaces $500 Monthly Software
Say goodbye to $500/month rendering software. Someone built a complete AI rendering studio that runs in your browser, costs nothing to start, and has 15+ specialized modules. It's called ArchiVinci. 740,000 architects are already using it. German firms use it. UK design studios use...
Blackstone and Google Launch $5 B AI Cloud Venture to Deploy Custom TPUs
Blackstone and Google announced a $5 billion joint venture to create a U.S.-based AI cloud company that will offer Google’s custom Tensor Processing Units as a compute‑as‑a‑service. The venture aims to bring 500 MW of data‑center capacity online in 2027, signaling deep...
Simulated Soulmates: How Common Are AI Companions?
A new study by BYU's Wheatley Institute and the Institute for Family Studies surveyed over 2,000 U.S. young adults in committed relationships and found that 15% regularly interact with AI romantic companions, dubbed “secret soulmates.” Between 20% and 30% have...
Amazon Names Veteran Matt Wood Chief AI and Technology Officer
Amazon has hired longtime AWS executive Matt Wood as its new chief AI and technology officer. The move underscores Amazon’s push to tighten AI product architecture and compete more aggressively with rivals like Microsoft and Google. Wood’s return comes as...
Exadel Acquires UK Design Firm Tangent to Fuse AI Engineering with Experience Design
Exadel, the Tampa‑based software services firm owned by Sun Capital, announced the acquisition of London‑based digital experience consultancy Tangent. The deal, terms undisclosed, integrates Tangent’s UX and MarTech expertise into Exadel’s AI‑native engineering practice, creating a single‑contract offering for enterprise...
Meta’s $200 B AI Data Center Redraws Rural Louisiana Real‑Estate Map
Meta Platforms is pouring more than $200 billion into Hyperion, a 5‑gigawatt AI data center in Richland Parish, Louisiana. The project will host up to 7,500 construction workers and reshape a swath of farmland the size of Manhattan, prompting a dramatic...
Legora and Datasite Launch Generative AI-Powered Due Diligence Platform
Legora announced a partnership with virtual data‑room provider Datasite to embed generative AI due‑diligence tools into the Legora platform. The integration lets users pull documents from Datasite VDRs and run AI‑driven analysis without leaving the workflow, preserving existing data permissions.
Intel and Qualcomm Eye Tenstorrent in AI Chip Consolidation Push
AI accelerator startup Tenstorrent has entered early-stage acquisition talks with Intel and Qualcomm. The interest reflects mounting pressure on major silicon vendors to secure more efficient AI processing technology and to counter Nvidia's dominance.
Google’s ‘Remy’ AI Agent Triggers Enterprise Security Concerns
A leak of Google’s internal Remy AI agent, a Gemini‑powered personal assistant, has enterprise architects questioning the security and integration of autonomous agents. Experts warn that long‑running workflows demand new infrastructure and governance, reshaping how enterprises will adopt AI stacks.
SurfaceGX Unveils AI Visibility Repair Platform to Fix Brand Misrepresentation in LLMs
SurfaceGX launched its AI Visibility Repair Infrastructure platform, a tool that not only monitors but also diagnoses why brands are missing or misrepresented in large language model answers and delivers concrete remediation. The solution targets the fragmented AI observability stack,...
AI Flood Overwhelms Linux Security Mailing List, Experts Warn of New Threat Landscape
Linus Torvalds warned that a surge of AI‑generated vulnerability reports has rendered the Linux security mailing list nearly unmanageable. The warning comes amid a wave of AI‑enabled supply‑chain attacks, including a massive npm compromise affecting millions of downloads, and growing...
Anthropic Adds Self-Hosted Sandboxes and MCP Tunnels to Claude Managed Agents
Anthropic has added self‑hosted sandboxes and Model Context Protocol (MCP) tunnels to its Claude Managed Agents platform. The sandboxes let customers run tool execution on their own infrastructure or via managed providers such as Cloudflare, Daytona, Modal, and Vercel, keeping...
How to Turn Off AI Notification Summaries for News Apps on Your iPhone
Apple has restored AI‑generated notification summaries for news and entertainment apps in iOS 26, the latest OS released in September 2025. The feature, previously pulled after a BBC report flagged inaccurate headlines, now appears with a beta disclaimer warning users to verify...
The Agentic Infrastructure Era
Pulumi announced a suite of new platform features designed to make infrastructure management fully agentic. By leveraging infrastructure-as-code in familiar languages, the company reports that LLM‑driven agents now handle about 20% of deployments, with a goal of exceeding 50% by...
LLMs Still Falter at 3D Vision Scientific Coding
#CVPR2026 Can frontier LLMs write PhD-level 3D vision code? We introduce GeoCodeBench, a benchmark that asks models to read real 3D geometric vision papers and implement core functions. Best result so far: GPT-5 reaches only 36.6%. This suggests that scientific coding in 3D...
Building Portable AI Workflows That You Can Take Anywhere
In this episode, host Michael Stelzner and AI expert Nicole Leffre discuss how marketers can build portable AI workflows that aren’t locked into a single platform. They debunk the myth that autonomous agents always outperform chat interfaces and warn against...
Why AI Needs an Intelligence Layer to Get Health and Wealth Right
The article argues that large language models alone cannot reliably answer employee benefits questions because the critical intelligence layer beneath the model is missing. It outlines four stacked layers—orchestration, data foundations, benefits structure, and claims intelligence—required to turn fragmented health...
Why an AI ‘Kill Switch’ Is Harder Than It Sounds
UK MPs have tabled an amendment to the Cyber Security and Resilience Bill that would give the Secretary of State authority to order the shutdown of data centres or AI systems in an emergency, though the proposal has not yet...
What If AI Didn’t Make Parenting Faster, But Made It Slower?
Playsense, a year‑old parenting app, has integrated ElevenAgents’ voice AI to let parents record observations by speaking instead of typing. The voice agent waits for pauses, mirrors tone, and extracts insights, turning casual ramblings into a structured child profile. The...
What Google’s UCP Tells Us About Agent-Ready Websites via @Sejournal, @Slobodanmanic
Google unveiled the Universal Commerce Protocol (UCP), an open‑source blueprint that lets AI agents discover and execute ecommerce actions via a standardized manifest and three checkout API calls. First rolled out for Google Merchant Center retailers, UCP defines a /.well-known/ucp...
Axios Harris Poll 100: GOP Embraces AI over Democrats
The latest Axios/Harris Poll 100 shows Republicans increasingly trust AI firms, while Democrats grow more skeptical. Republican favorability toward AI rose to 44% versus 35% for Democrats, and the partisan gap for OpenAI widened from 1 point in 2024 to...
How AI Is Transforming Network Incident Response (and Where It Still Falls Short)
AI is reshaping network incident response, but its impact is bounded by a severe visibility gap. While 95% of IT professionals admit they cannot see large portions of their networks—especially in public cloud—AI can only act on the data it...
An AI Data Center in Your Home?
Resistance to large AI data centers is prompting pilots for residential‑scale compute. Companies like PulteGroup, Nvidia and Span are exploring basements, utility rooms, and detached structures as micro‑data centers. Homeowners see potential income to offset mortgages, while businesses seek lower‑cost,...
Context Graphs and Decision Traces to the Rescue
Foundation Capital’s December 2025 paper introduced “context graphs,” a knowledge‑graph layer that records AI decision traces—how rules were applied, exceptions granted, and approvals made. By capturing episodic, semantic, and procedural memory, context graphs aim to give enterprise AI agents explainable,...
AI Post-Slicing GCODE Lowers FFF Material Use
Researchers at King Saud University have demonstrated an AI‑driven post‑slicing system that trims fused filament fabrication (FFF) material consumption by roughly 22‑25% without altering CAD models or slicer settings. The framework adds curvature‑aware adaptive layers, function‑specific reinforcement, and unsupervised weak‑zone...
Anthropic Denies EU Access to Claude Mythos, ChatGPT 5.5 Comes to Rescue
Anthropic has refused EU requests for access to its Claude Mythos model, despite the model’s claim of uncovering thousands of high‑severity vulnerabilities. OpenAI stepped in, offering the EU its ChatGPT 5.5‑Cyber model, which the UK AI Security Institute says matches or exceeds...
Alibaba Teases New Qwen Previews, Highest-Ranking Chinese AI Models on Arena
Alibaba unveiled preview versions of its next‑generation Qwen 3.7 models—Max and Plus—ranking 13th in text and 16th in vision on LM Arena. The previews follow the Qwen 3.6 launch and precede a full release at the Alibaba Cloud Summit in Hangzhou. AI‑related revenue...
Tabular ML Is Entering a New Benchmark Era
Tabular machine learning is moving from static paper‑based benchmarks to live, capability‑oriented leaderboards such as TabArena, ScoringBench, and MulTaBench. The rise of tabular foundation models like TabICL and TabPFN‑3.0 enables a single pre‑trained model to be evaluated across tasks—classification, regression,...
Structured Harness Turns Claude Code Into Reliable Engineer
I don’t trust Claude Code without this anymore. A developer built a free open-source harness that forces Claude Code to stop acting like a chaotic intern and start working like an actual engineer. It’s called Claude Code Harness. Most AI coding sessions fail...
China Calls for Integration of AI, Advanced Manufacturing Amid Economic Headwinds
Chinese Premier Li Qiang urged deeper integration of artificial intelligence with advanced manufacturing to create new growth momentum amid slowing domestic demand and rising external pressures. During visits to Xiaomi’s electric‑vehicle plant and the Humanoid Robot Innovation Centre, he called...
How Sulfman’s Eagle Watch Is Putting African Innovation on the Global Map
Sulfman Consulting, a Nigerian cybersecurity firm founded in 2015, launched Eagle Watch, an AI‑powered digital investigation platform, at GITEX Nigeria in September 2025. The tool leverages open‑source intelligence to provide affordable, token‑based searches for due‑diligence, fraud prevention, and risk profiling across...
AI Coding Made Us Faster. Why Did Incidents Increase?
Engineering teams that adopt AI‑coding assistants see immediate gains in velocity—larger pull requests, shorter cycle times, and faster sprint completions. However, DORA data shows that the same acceleration often leads to a higher change‑failure rate, causing more hot‑fixes and on‑call...
Why Price Transparency Took a Decade to Crack | Heather Fernandez, Solv
In this episode, Heather Fernandez, CEO of Solve, discusses the decade‑long struggle to achieve price transparency in U.S. healthcare and how her company is finally cracking the "how much will it cost me?" question using AI-driven ClearPay. She traces her...
Americans’ AI Hate Wave Might Just Be Gathering Steam: Data Centers Could Hike Power Costs in some States over 50%...
The share of U.S. electricity used by data centers more than doubled from 1.9% in 2018 to 4.4% in 2023, and analysts project it could reach as high as 17% by 2030. This surge could push wholesale power prices up...
Why ‘Open AI’ Models Are Gaining Ground on LLMs
Open‑weight AI models are gaining traction as enterprises seek greater control, lower costs, and the ability to fine‑tune models for specific workloads. While proprietary LLMs like ChatGPT and Gemini dominate headline usage, open models such as Meta Llama, Mistral, DeepSeek...
XPENG Launches Mass-Produced Robotaxi in China With L4 Autonomy
XPENG has begun mass production of its first robotaxi in Guangzhou, marking the first domestic L4‑level autonomous vehicle built entirely with in‑house hardware and software. The vehicle runs on the GX platform and is powered by four proprietary Turing AI...
AI Addiction Emerges as Major Business Liability Risk
Addictive AI Could Become The Next Big Business Risk #AI is making digital products more engaging, but that also creates a growing #risk around #addiction , #mentalhealth and harmful customer behavior. As regulators, courts and consumers pay closer attention, #businesses...
GitHub Makes GPT-5.3-Codex the Copilot Base Model for Enterprises
GitHub announced that, effective May 17 2026, GPT‑5.3‑Codex will replace GPT‑4.1 as the default base model for Copilot Business and Enterprise. The new model is designated as a Long‑Term Support (LTS) release, with guaranteed availability through February 4 2027, easing compliance and security reviews....
Generative AI Is Changing the Game in Video Gaming
Ubisoft revealed its new generative‑AI middleware that powers AI teammates capable of listening and responding in real time. The system relies on small, on‑device language models, advanced prompt optimisation, and low‑latency inference to deliver immersive, context‑aware interactions without heavy cloud...
Meta Is Reportedly 'Reassigning' 7,000 Employees to AI-Focused Roles
Meta announced a major internal shift, moving 7,000 employees into four new AI‑focused organizations while simultaneously laying off 8,000 staff, about 10% of its workforce. The reassignments aim to create flatter structures and “AI‑native” design processes, with employees told to...
Agent Success Hinges on Precise, Well‑structured Data Context
This is true of all agents, not just coding agents. Probably the biggest challenge that most companies run into in their agent strategy is getting agents the right constrained context to work with for a task. Too much information or...
Google, Blackstone Create AI Cloud Company
Google and Blackstone announced a joint venture to launch an AI‑focused cloud company, backed by a $5 billion equity commitment. The new entity will deploy roughly 500 MW of AI compute capacity by 2027, leveraging Google’s Tensor Processing Units and Blackstone’s digital‑infrastructure...
Never Switch AI Models Mid‑Workflow—Lose Your Context
Advice for using Claude Code or Codex: don’t switch models mid-workflow. A model switch is like handing the task to a new engineer with the same laptop, files, terminal history, and notes; but none of the previous engineer’s mental flow. Anything...
ChatGPT Images 2.0 Turns Text Into Stunning Visuals
ChatGPT Images 2.0 is just freaking good. All I did was drop in the raw text from my @OutSystems article + their logo... and it outputted this 🤯 https://t.co/cGUHsHKiY1
Tracking Difficulty with Feature Portfolios
The article argues that forecasting AI capabilities requires task attributes that are measurable, interpretable, stable, and sufficiently explanatory. It shows that human completion time, the traditional metric for time‑horizon forecasts, is increasingly inadequate as tasks grow longer and more complex....
AI Grid Superiority Drives National Prosperity
Compute is the most critical infra for the USA. The countries with the best AI Grids will prosper the most.
PwC Deploys Anthropic’s Claude AI to 75,000 Staff for Deal Execution
PwC announced it has rolled out Anthropic’s Claude large‑language model to 75,000 employees worldwide, making it the firm’s default AI tool for deal execution. The deployment, the largest of its kind in a professional‑services firm, aims to accelerate transaction analysis,...
GSA's OneGov AI Program Reaches 3.4 Million Federal Users, Unlocking $1.15 B in Savings
The General Services Administration announced that its OneGov AI buying program now provides AI tools to roughly 3.4 million federal users after more than 120 orders, generating $1.15 billion in identified savings. The move accelerates AI adoption across dozens of agencies while...