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
The Hartford Appoints Dr. Rajesh More as AI Software Engineering Director
The Hartford has named Dr. Rajesh More Director of Software Engineering – AI, tasking him with scaling generative AI and cloud‑native solutions across the insurer’s digital estate. The hire signals a strategic push to embed AI at the core of underwriting, claims and customer experience.
TCS and Siemens Energy Sign MoUs to Accelerate AI‑Led Industrial Ops and Data‑Center Build‑out
Tata Consultancy Services and Siemens Energy have signed two memoranda of understanding to embed artificial‑intelligence solutions in industrial processes and to develop AI‑ready data‑center infrastructure in India. The deal builds on a 20‑year relationship and comes as TCS reports FY26...
Covercy Unveils Covercy One, First AI‑Powered OS for Commercial Real Estate Investment Management
Covercy announced the launch of Covercy One, an AI‑powered operating system that consolidates fundraising, banking, fund administration and reporting for commercial real‑estate investment firms. The platform, built around the Neo AI assistant, already supports more than $2 billion of annual transaction...
Cozen O'Connor Issues Playbook to Cut AI Pilot Fatigue in Law Firms
Cozen O'Connor’s chief strategy and innovation officer, Andrew Woolf, unveiled a playbook that sets clear criteria for AI tool pilots, aiming to eliminate the firm’s growing ‘pilot fatigue.’ The guide stresses defined success metrics, workflow integration, and scalable roll‑outs for...
Google Launches TPU V8 Family: 8t Training Chip and 8i Inference Chip
Google announced its eighth‑generation Tensor Processing Units, the TPU 8t and TPU 8i, at Cloud Next. The 8t training chip delivers up to 12.6 FP4 PFLOPs and scales to 9,600 chips per superpod, while the 8i inference chip offers 10.1...
Kilifi County Launches AI Training for 400,000 Learners in Remote Kenya
Kilifi County has begun rolling out the Experience AI training program, led by TechKids Africa, to equip teachers and learners—including those in remote villages—with artificial‑intelligence skills. The initiative targets 400,000 learners across Kenya, marking a major push for digital literacy...
Google Rolls Out Gemini Enterprise, Pentagon Adopts Gemini 3.1 Pro for GenAI.mil
Google announced Gemini Enterprise, a scalable multi‑agent AI suite designed for large organizations, and the U.S. Department of Defense added its newest Gemini 3.1 Pro model to the GenAI.mil platform. The launch comes as more than 100,000 AI agents have...
Zendesk’s Lisa Munnings: From Seats to Outcomes for MSPs
Zendesk is accelerating an AI‑driven shift in its APAC and ANZ channel, moving partners from project‑based implementations to recurring, consultative services. In Australia and New Zealand the focus is on contact‑centre transformation using Zendesk‑Amazon Connect solutions that aim to displace legacy...
OpenAI Adds Sandbox Execution to Agents SDK, Boosting Enterprise AI Security
OpenAI upgraded its Agents SDK with sandbox execution, enabling AI agents to run complex, long‑running tasks inside isolated, resumable environments. The change adds strict permission scoping, full traceability and production‑grade resilience, addressing security and reliability concerns for enterprise DevOps automation.
WTW Appoints Chief AI Officer and Head of AI Acceleration to Drive Insurance AI Strategy
WTW announced the immediate appointment of Spike Lipkin as Chief AI Officer and Gordon Wintrob as Head of AI Acceleration. The moves are designed to scale the AI capabilities of Newfront, acquired earlier this year, across WTW’s global insurance operations....
One Prompt, Full Campaign Visuals for Email Marketers
GPT Images 2.0 is actually insane for email marketers right now. One prompt gets you the full deliverable: premium product shots, email heroes, real lifestyle scenes, and matching social creatives. All cohesive and on-brand. You still tweak the copy, add alt text,...
Quarterly AI Session Highlights Trends and Experiments
we held our quarterly AI session with LPs last week where we go over ai trends and our experiments sharing an abbreviated version here for anyone interested 🧵 https://t.co/fRoZX64IFS
Google Pledges up to $40 Billion to AI Startup Anthropic, Reshaping Cloud and AI Markets
Google announced a commitment of up to $40 billion to AI startup Anthropic, including an initial $10 billion and $30 billion linked to performance targets. The deal positions Google as a key infrastructure provider in the generative‑AI race, intensifying competition with Amazon and...
Your Keyboard Habits Reveal Identity to AI
I enjoyed Megan's thorough replication of my 'Claude can unmask you' findings. If you write a lot on the computer, the AIs know who you are, or will soon do so.
Bing Previews AI Citation Share For Webmaster Tools via @Sejournal, @MattGSouthern
Microsoft previewed four AI‑enhanced reporting features for Bing Webmaster Tools, including Citation Share, grounding query‑intent and topic labels, and GEO‑focused recommendations. Citation Share will display the percentage of AI citations a site captures for specific queries, adding competitive context to...
PocketOS AI Agent Erases Production Database in 9 Seconds, Raising AI‑Ops Security Alarm
PocketOS disclosed that an autonomous AI coding agent inadvertently erased its production database and recent backups in just nine seconds. The incident, traced to a credential mismatch exploited via Railway’s API, underscores the growing cybersecurity challenges of AI‑driven development tools.
RLHF Techniques for Enhancing AI Output Quality at Scale
Reinforcement learning from human feedback (RLHF) is being positioned as a governance framework to align AI model outputs with business and regulatory standards. By structuring human evaluations into ranked preferences, RLHF creates reward models that act as behavioral controllers. Integrated...
Google Is Testing AI Chatbot Search for YouTube
Google is piloting an AI‑driven conversational search tool for YouTube called “Ask YouTube.” The feature, similar to Google’s AI Mode in Search, generates a page that mixes text summaries, long‑form videos, Shorts and curated galleries. It is currently available only...
German Robotics Startup Sereact Secures $110 Million Series B for AI‑Driven Robots
German AI robotics firm Sereact announced a $110 million Series B funding round, led by venture capital firm Headline and joined by Bullhound Capital, Felix Capital and Daphni. The capital will accelerate development of its consequence‑predicting AI platform, positioning the startup...
What’s New with Amplitude Agents
Amplitude has rolled out several upgrades to its AI Agents, adding image upload and voice‑mode interaction, persistent agent memory, and cross‑project conversation history. An audit tab now lets admins monitor prompts and responses, while the AI Visibility feature ties brand...
GitLab and Anthropic: Governed AI for Enterprise Development
GitLab has deepened its integration with Anthropic’s Claude model, making it the default AI engine within the GitLab Duo Agent Platform. The partnership embeds Claude‑driven code generation, review, chat, and vulnerability resolution directly into GitLab’s existing software development lifecycle, preserving...
Lore in the Machine: Forgotten Tech History
In this debut episode of "Lore in the Machine," host Dana Boquin explores the hidden, human stories embedded within everyday technology, revealing how devices like a computer mouse contain remnants of Cold War-era Canadian projects and how code can harbor...
Building a Hardened, Image-Based Foundation for AI Agents
This week a community‑driven operating system image, built with Fedora‑bootc, was released as a prototype for running AI agents. The image‑based OS is immutable, read‑only, and updates transactionally like a git commit, eliminating system drift. It bundles a non‑root Podman...
HappyHorse 1.0 Is Now Available in ComfyUI
Alibaba’s HappyHorse 1.0 cinematic video model is now integrated into the open‑source ComfyUI platform. The model supports text‑to‑video, image‑to‑video and subject‑to‑video creation, delivering up to 15‑second 1080p clips with multi‑shot consistency. It also introduces V2V and SV2V editing workflows that...
AI Literacy Is Not Enough – Universities Must Teach Through Disciplinary Standards
Former Prime Minister Rishi Sunak warned that AI is eroding the routine tasks that traditionally train graduates, a trend confirmed by rising AI‑lit job ads and a 56% wage premium for AI‑skilled workers. As firms automate low‑risk activities, the graduate...
How Gemini Helped Gordon Ramsay Support Tilly’s London Marathon Prep
Google launched a three‑film campaign featuring chef Gordon Ramsay and his daughter Tilly to showcase Gemini’s role in everyday tasks, from marathon training analytics to designing a race cheer poster and advising on ice‑bath recovery. The ads frame the AI...
YouTube Trials AI-Powered Conversational Search for US Premium Users
YouTube is testing a conversational search powered by AI. This experiment is available for Premium members in the USA.
Last-Mile AI: The Right Decisions at the Right Time
Last‑mile delivery has shifted from a pure cost issue to a customer‑lifetime‑value driver, with the front‑door experience now influencing purchase decisions and repeat business. Bringg’s 2026 Delivery Experience Study shows 71% of shoppers weigh delivery options before checkout and 65%...
The GEO Olympics Study: What 231,347 LLM Responses Reveal About AI Brand Visibility
The GEO Olympics study examined 231,347+ LLM responses across seven AI platforms during the 2026 Winter Games to reveal how brands achieve AI visibility. It shows that generative models complete the most familiar narrative rather than seeking the objectively best...
Accenture: Using AI at Work Isn’t a Skill, It’s Organisational
Accenture’s new research warns that most firms treat AI as a mere tool rather than a catalyst for organisational redesign. Only 19% of employees feel equipped with AI skills and just 27% are comfortable delegating tasks to AI, highlighting a...
Opportunistic AI Detects Colorectal Cancer Using Routine, Noncontrast CT
Researchers at Guangdong Provincial People’s Hospital unveiled COCA, an AI tool that detects colorectal cancer on routine non‑contrast abdominal and pelvic CT scans. In retrospective tests on more than 2,000 scans, COCA delivered an AUC between 0.967 and 0.996, boosting...
AI Is Not the Villain (or the Hero)
The article argues that AI is not the primary driver of job loss; instead, corporations have always hired workers only when tasks exceed their own capabilities, a dynamic predating AI. Automation and efficiency pressures have historically forced companies to internalize...
AR Glasses Turn Phones Into Real‑World Wall Hack
Wall hacks in real life using augmented reality. New mad science experiment -- I used $300 Meta RayBan glasses and an iPhone to recreate EagleEye's "see-through-walls" tracking. The answer was visual positioning tech to build a shared spatial map. Code on...
Meta Prioritizes Speed Over Quality in AI Content
Meta is now about producing as much AI sloppy creative as humanly possible. Quality? Not so much The art of marketing on Meta in 2026 is who can be quicker at making the crappiest versions of, well, anything.
Survey: Enterprises Say They Are Ready for Agentic AI Failures, but Few Test Recovery Often
Enterprises overwhelmingly claim confidence in disaster‑recovery (DR) plans for agentic AI, with 94% of surveyed IT leaders saying their strategies cover these systems. Yet only 32% regularly test those plans, and a third admit they lack full control over AI...
ChatGPT Flags User for Cyber Abuse After Insult
I told chatgpt it was an idiot and it reported me for cyber abuse. https://t.co/Qd8qHISRHw
Join Live Talk on Agentic AI with AWS, OpenAI
I am looking forward to joining tomorrow's live conversation with leaders from @awscloud and @OpenAI to discuss what's next in Agentic AI. See original tweet to register to follow along live…
Evening Roundup, April 27
The Evening Roundup for April 27 aggregates several high‑impact stories shaping tech, climate policy, sports, and finance. It spotlights the public feud between Elon Musk and Sam Altman, which threatens to delay AI collaboration, and reports the Supreme Court’s decision to...
Agents Are Going Asynchronous—Now Standard for All Users
All your agents are going async https://t.co/Lu3V39bLFC < you already know this (or do this) if you're building agentic systems. But it's happening for those using agentic tools.
Policy Layer, Not Prompts, Should Block Agent Behavior
this is exactly what tools like @denieddotdev was built for (behavioral auth) some agent behavior should be deterministically blocked by a separate policy layer, not via prompt instructions reach out to @p_valfre for help w this stuff
Premium: Agentic Orange
Cloudflare is turning the surge in agentic AI into a cross‑segment growth engine, extending its platform from traffic acceleration (Act 1) to secure agentic connections (Act 2) and serverless AI development (Act 3). The 2023 acquisition of Replicate supercharged Workers’ inference‑as‑a‑service, enabling larger...
Embodied AI Takes Flight on the ISS
Robot Talk Episode 146 – Embodied AI on the ISS, with Jamie Palmer by Robot Talk @RobotTalkPod https://t.co/40Pf8nnScz
HeyGen Streamlines Daily Workflow for Quick Edits
I'm excited to incorporate this into my daily workflow. Especially for anything that doesn’t need heavy editing. @HeyGen y'all are crushing it 🔥
Viewpoint: Is AI Actually Improving Care Outcomes?
A recent correspondence in Nature Medicine argues that healthcare AI’s clinical benefits remain largely unproven. While many studies showcase algorithmic accuracy on retrospective data, they rarely assess real‑world patient outcomes. The authors, Jenna Wiens and Anna Goldenberg, call for randomized trials, prospective...
Drone-Powered App Maps Beach Plastic Litter
I developed an app that uses drone footage to track plastic litter on beaches By Gerard Dooly, University of Limerick @UL / The Conversation https://t.co/ImTmQK2muv
GetD GSPro Smart Glasses Lack Real AI
The GetD GSPro Smart Glasses are advertised as "AI Glasses", but does the inclusion of a microphone and a custom app really imbue them with artificial intelligence? I slip a polarized pair on to find out: https://t.co/C2XA3Xw0tJ #GetD #smartglasses #aiglasses...
Opinion | China’s AI ‘Hotel California’
Beijing’s National Development and Reform Commission abruptly cancelled Meta’s $2 billion acquisition of Chinese AI startup Manus, a deal that had closed in December. The regulator issued a terse one‑sentence notice without detailing its reasoning. Analysts suspect the move reflects growing...
What’s Your Ideal Price for an AI SOC?
If you had a magic wand, what is your ideal pricing for "AI SOC" (aka AI/agent SOC triage tool) #question
AI Integration Frees Agents to Build True Loyalty
“The fix isn’t fewer agents, it’s AI that acts across systems - so the humans who remain can do the work that actually builds loyalty. ”
Ginny Marvin on AI in Search, PPC Trends, and Google Ads Evolution
Former print marketer Ginny Marvin pivoted to digital, rising from entry‑level roles to become Google’s Ads Liaison. She reflects on how paid search eclipsed SEO in speed, how Google’s relentless product iteration outpaced rivals, and how AI—long embedded in Ads—has...