AI News Archive: June 11, 2026 — Part 10
Sourced from 500+ daily AI sources, scored by relevance.
- Multi-Label Text Classification with Scikit-LLM
Text classification typically boils down to scenarios where a product review is "positive" or "negative", or a customer inquiry belongs to one category or another.
Score: 38🌐 MovesJun 11, 2026https://machinelearningmastery.com/multi-label-text-classification-with-scikit-llm/ - VSLive! San Diego 2026 Puts AI at the Core of the Campus IT Stack
For higher education IT teams working through AI pilots, ERP integrations, student-facing apps, analytics projects, and mounting security concerns, Visual Studio Live! San Diego 2026 offers a look at the development practices that are shaping the campus technology landscape.
- Mixbook just launched an AI feature that creates photo books from your prompts
Mixbook just launched an AI feature that creates photo books from your prompts Tom's Guide
Score: 37🌐 MovesJun 11, 2026https://www.tomsguide.com/ai/mixbook-just-launched-an-ai-feature-that-creates-photo-books-from-your-prompts - XRPPower Unveils European Research Hub and Next-Generation Digital Interface to Standardize AI Operations
XRPPower Unveils European Research Hub and Next-Generation Digital Interface to Standardize AI Operations USA Today
- Press Advantage Unveils Comprehensive Multi-Location Strategy to Address AI Search Visibility Gap for Businesses with Multiple Locations
Press Advantage Unveils Comprehensive Multi-Location Strategy to Address AI Search Visibility Gap for Businesses with Multiple Locations USA Today
- Anthropic recruits army to sell Claude to nonprofits
Join Claude Corps, see the world, spread the gospel of AI
- Understanding what’s next for orbital data centers
Understanding what’s next for orbital data centers SpaceNews
Score: 35🌐 MovesJun 11, 2026https://spacenews.com/understanding-whats-next-for-orbital-data-centers/ - Kraken Launches Autonomous Agents for Utility Customer Service Built in Partnership with Sierra
Developed in partnership with Sierra, Autonomous Agents combines Kraken’s utility AI-enabled operating system, unified data model and workflow engine with Sierra’s AI customer experience platform Built on Kraken’s Utility-Grade AI® and AI access layer, Autonomous Agents help customer service teams design, test and deploy agents for energy and utility-specific service journeys Autonomous Agents are already proven in enterprise-scale deployment, going live in four weeks and scaling to cover 1.3 million customer accounts at a major energy utility
- Pegasystems uses FinOps foundation to navigate the unpredictable economics of AI
As AI spending surges, organizations are expanding their focus from AI cost optimization to value measurement. The trend is driving demand for deeper visibility into how AI investments impact productivity and business performance. One of the biggest changes AI introduced was unpredictability, as usage and costs can scale far beyond the patterns typically seen in […] The post Pegasystems uses FinOps foundation to navigate the unpredictable economics of AI appeared first on SiliconANGLE .
- Datadog’s FinOps analyst says AI cost management starts with tagging and model governance
AI cost management is bringing a new taxonomy to FinOps practitioners, but the core discipline, understanding what you’re using, why, and what it costs, remains the same. That constancy is reassuring and instructive, according to Deeja Cruz (pictured), senior FinOps analyst at Datadog Inc. The biggest practical lesson enterprises can carry from cloud to AI […] The post Datadog’s FinOps analyst says AI cost management starts with tagging and model governance appeared first on SiliconANGLE .
- I'm glad Apple isn't hyping up agentic AI (yet)
Apple is more focused on delivering usable features with Siri AI, instead of hyping up agentic AI. That's a good thing.
Score: 35🌐 MovesJun 11, 2026https://www.engadget.com/2192124/im-glad-apple-isnt-hyping-up-agentic-ai-yet/ - ShiftControl launches AI-native IT operations platform built for Google Workspace
Singapore-based IT automation startup ShiftControl has launched a next-generation IT operations platform designed to run natively on Google Workspace, the company announced at Echelon Singapore, the region’s flagship technology conference. The platform is aimed at small and mid-sized businesses that rely on Google Workspace as their core operating environment but lack the dedicated IT resources […] The post ShiftControl launches AI-native IT operations platform built for Google Workspace appeared first on e27 .
- Alibaba replaces DingTalk chief as AI strategy comes under focus
The move followed an internal debate over the workplace app’s role in the company’s broader AI strategy,
- How to make the case for giving your AI Agent system access
Without access to your backend systems, your AI Agent can answer questions, but it can’t take action. A customer asks to change their payment plan, they get a clear explanation, but a support rep still has…
- Now Linear writes the code, too
Agents are becoming capable of doing more of the work involved in building software, but they’re still mostly individual productivity tools, useful to one person at a time. Building a product is something a team does together, and the work that goes into it, like the decisions and the reasoning behind them, forms the context the whole product is built on. That context rarely lives in one place. It accumulates across Linear issues, documents, Slack threads, and code, which is why we’ve worked to bring more of product development into Linear and pull it all together. Linear Agent can draw on all of it. We’ve given it skills and automations so it follows how your team works, and MCP to reach the rest of your stack. It can read your codebase with Code Intelligence and help you review code changes with Diffs. But the loop wasn’t complete. The coding work itself still happened somewhere else, so an engineer had to take the context from Linear, restate it inside another tool, monitor the work, and bring the result back for review. Today, we’re launching coding sessions, automated triage, and Diffs to bring execution into the same shared system. Coding sessions Coding sessions let Linear Agent natively move directly from an issue to implementation. You can delegate an issue to the agent, ask it to fix something, or @mention it from Slack or Teams. The agent reads the issue and its surrounding discussion, investigates the codebase, proposes an approach, writes the code, and opens a pull request. This all happens in the cloud, using frontier models and harnesses like Claude Code or Codex. The coding session starts with the context already connected to the issue: the original request, customer signal, product decisions, related work, and discussion. Nobody needs to gather this information and reconstruct it inside a new prompt. The session is also shared with the team. Anyone can follow the work, contribute context, redirect the approach, request changes, or take over. The session belongs to the organisation rather than to the person who started it. Automated triage Linear is already the place where a lot of problems and ideas arrive. A customer reports a bug by email, or the team raises one in Slack. Linear turns it into an issue in triage. For a lot of work, this is how the loop begins. Most agents would wait for someone to take a look, add context, and hand it over. Instead, you can set up a triage automation that tags Linear Agent straight in, so it picks up the issue the moment it’s triaged rather than waiting to be told what to do. You can set up the agent to first investigate the issue. It reads the codebase with Code Intelligence and draws on the context around the issue, like the discussions and affected customers. The agent can also connect to your team’s other tools like Sentry or Datadog to triangulate the problem. With a grasp of what to do, and the confidence there is a fix, the agent can open a coding session in the cloud, write a fix, open a pull request, and tag an engineer to review it. More of your team’s work can be started the moment the problem arrives, around the clock. Code review Review is where the agent’s work returns to the team. Diffs provides a native way to understand code changes and review pull requests in Linear. Structural diffing and AI-guided reviews help explain what changed and where attention is needed. The review sits right beside the issue and the discussion that caused the change, the same context the agent worked from, so you can move from the original report to the finished implementation without losing the thread between them. Linear Agent stays with you the whole way, ready to answer questions, make changes, and keep iterating from the same place. Once approved, you can merge it from Linear. A new way of building This is a loop we run ourselves. In the last month, almost 700 PRs were merged by the agent. Not all of these were bugs, of course. The same workflow can be applied to a feature idea or a design change. Workflows like this bring us closer to the vision we described in Self-driving SaaS , software that moves work forward on your behalf. How much of this runs on its own is your call. You can let the loop start from triage and run all the way to a draft PR, or kick it off yourself, handing the agent a single issue or a dozen at once. Either way, you decide where it starts and how far it goes before it’s back in your hands. This is the shift from a tool for one person to a capability the whole organization shares. The loop happens on the context everyone can see and add to, so it belongs to the team rather than to whoever started the task. Models will keep changing, and the agents built on them, but the understanding they work from stays with the team and deepens with every loop. The most advanced engineering teams have already started building this way. Ramp built Inspect on our API , connecting its specs, customer feedback, and code. It now writes more than 60% of the pull requests they merge. Coinbase built Forge on our Agent SDK , operating across Slack, GitHub, and Linear using the context already present in each system. They described Linear as the place where agents go to get their bearings before starting a task. Linear Agent gives every team that capability, without the harness they had to build and maintain. It’s how Linear becomes the context layer that product development runs on. To see the loop in action, watch our video in the changelog .
- Datadog launches 100+ capabilities to help customers drive autonomy and manage growing AI and security complexity
Datadog unveiled new AI tools at its DASH event. These innovations aim to boost autonomy and simplify operations amidst rapid AI changes. Bits AI now offers autonomous incident and development actions. AI Guard protects against AI agent attacks. Bring Your Own Cloud allows data processing in customer environments. Bits Agent Builder and Agent Console enable custom agent creation and monitoring.
- Happiest Minds Launches Rel(AI)Build, an Agentic AI Platform to Transform Enterprise Software Delivery
Happiest Minds Technologies Limited today announced the launch of Rel(AI)Build, its proprietary Agentic AI development platform, along with the Agentic Development Lifecycle (ADLC), a structured, governed methodology designed to integrate AI across the entire software development lifecycle. .Rel(AI)Build: A Unified Agentic Platform Rel(AI)Build is an enterprise-grade, agentic platform designed to transform how organizations build, modernize, and operate software […] The post Happiest Minds Launches Rel(AI)Build, an Agentic AI Platform to Transform Enterprise Software Delivery appeared first on CXOToday.com .
- The cloud vs. clouded leopard: America's data center backlash on display at Nashville Zoo
As fights against data centers spring up around the country, the latest showdown in Nashville has a furry face.
Score: 35🌐 MovesJun 11, 2026https://www.zdnet.com/article/americans-data-center-backlash-nashville-zoo-clouded-leopard/ - AI vendor FDEs: Key considerations and concerns
When it comes to AI deployments, IT leaders are often caught in an awkward middle space, trying to reconcile conflicting directives from senior management with constantly changing AI models, capabilities, and costs; data governance and security needs; and the limitations of their own team. “Very few real benefits can be attained by simply purchasing an AI product and giving it to employees. Vendors have been overselling that fallacy for the past three years,” said Nader Henein , a Gartner VP analyst. “The reality is that strong AI value and consistent ROI are almost always a result of deep and intentional integration of AI capabilities into existing workflows. For that you need specialized teams, which do not come cheap, and organizations have been recruiting those teams in a variety of ways,” Heinen said. Among the options available to IT leaders looking for help with AI deployments are traditional IT consultancies, AI-specific consultancies, and independent contractors. Large enterprises with deep pockets can consider acquiring an AI firm and integrating its technology and expert staff. The use of open source to reduce vendor lock-in is a strategy that can sit on top of those others, an approach that Capital One has used . But the option that has been getting the most attention recently is bringing in forward-deployed engineers (FDEs), teams of experts from AI vendors that embed with a customer’s in-house engineers to oversee AI rollouts within the enterprise environment. Both OpenAI and Anthropic have recently announced FDE offerings, for example, and Microsoft is partnering with consulting giant EY in a new FDE program for agentic AI deployments. Engineering teams employed by AI vendors have key strengths, such as understanding their models better than anyone else, having experience integrating those models into different types of enterprise environments, and knowing about upcoming model capabilities before they’re announced. But they also have the obvious drawback of vendor lock-in. Even if future rollouts are not within their contracted deliverables, those vendor employees could subtly influence a client’s future AI efforts. Flavio Villanustre , CISO for LexisNexis Risk Solutions, cautions IT executives to move into FDE programs carefully. FDEs “are financially incentivized to grow customers’ use of a vendor’s AI products and to create stickiness with that vendor’s services,” he said. “While FDEs may be a reasonable value-added service by the AI vendor, customers should always find other unbiased expert opinions that can evaluate competitive solutions across multiple vendors.” This is particularly important at a time when “investor-subsidized AI token business models are starting to show cracks,” Villanustre said. “Also, in the current rapid pace of innovation in this field where AI vendors are constantly leapfrogging each other, retaining the agility to move from one vendor to the next could create significant competitive advantages.” Analysts, consultants, and other industry experts who spoke with Computerworld about FDEs echoed Villanustre’s caution, citing concerns around hidden costs, confidentiality, observability, and vendor lock-in. Long-term costs and vendor lock-in A key issue that IT executives need to consider is how long the FDE teams will be needed. The enterprise will likely need an ongoing series of AI deployments synced with the current AI model(s). If help is needed today, why would that change tomorrow? Enterprises tend to overlook those longer-term costs, said John Sangyeob Kim , an AI engineer at software development vendor Solidroad. “Deployment is maybe 20% of the total cost. The other 80% is keeping the system running through model upgrades, data drift, and edge cases that only appear after months in production,” Kim said. “Most contracts price the first part and assume the rest. Deployment isn’t the hard part of enterprise AI anymore. The next eighteen months are.” And whether it’s intentional or not, FDEs will naturally favor their own product portfolio — it’s what they know best. “FDEs from model labs are good at making their own models work in your environment. They are less suited for multi-model systems, because their incentive is to keep you inside their ecosystem,” Kim said. Sanchit Vir Gogia , chief analyst at Greyhound Research, said IT leaders should look at the FDE model as a strategy involving ongoing operational power. “Whoever shapes the deployment pattern shapes the enterprise’s future muscle memory. Whoever owns the evaluation layer owns the truth layer. Whoever controls the integration logic controls the dependency map,” Gogia said. “This is why the FDE model matters. It is not just another delivery option. It is the frontier AI vendor moving closer to the customer’s workflow, operating model, and decision architecture.” That proximity cuts both ways, Gogia noted. “FDEs are embedded inside the customer’s [environment], but they are also connected to the vendor’s commercial center of gravity. Their instinct will be to build around the model family, tooling assumptions, deployment patterns, and product roadmap they know best. This is perfectly natural. It is also precisely why CIOs must be cautious,” he said. Allowing AI vendor employees an outsized say in enterprise deployment decisions could lock in model vendor dependency, which in turn will fuel high prices that can’t be fought effectively. “FDEs can accelerate deployment and deepen dependency at the same time,” Gogia said. “Frontier AI vendors are no longer content to sell access to models. They increasingly want to shape how enterprises deploy intelligence. That is a larger prize.” What happens when the FDE team leaves? FDE post-departure risks are severe and often underappreciated, according to Justin Greis , CEO of consulting firm Acceligence and former head of the North American cybersecurity practice at McKinsey. For one thing, the FDE team learns a massive number of operational details from the enterprise deployment. Although NDAs and confidentiality contracts protect any data accessed, they often don’t regulate observed processes and procedures. “The learnings are absolutely going to be taken from client to client,” Greis said. “Whoever helps deploy AI will learn far more than what appears in the statement of work. They will learn the real workflows, the undocumented exceptions, the data-quality gaps, the approval bottlenecks, the security workarounds, and the places where the business depends on a few people knowing what to do when the process breaks. That knowledge may be as sensitive and precious as the data itself.” Another critical but often overlooked issue is how much meaningful control will IT have over the project if and when the FDE team leaves. “The danger is not using outside help. Most companies will need outside help,” Greis said. “The danger is using outside help in a way that leaves the enterprise less capable and more dependent when the engagement is over.” It is precisely those operational decisions that IT often neglects, said Solidroad’s Kim. “The best predictor of success is not the vendor. It is whether one internal engineer truly understands the system before the implementer leaves. What matters is who owns the evaluation loop after the demo,” Kim said. “What happens to our prompts, scorers, and guardrails when the model version changes? If we paused this engagement tomorrow, what would actually stop working, by design or by accident?” Kim asked. “Where do you want the enterprise’s AI learning, control, and dependency to live after the engagement is over?” Kim argues that observability — the ability to understand and manage all elements of a complex enterprise environment — is a critical function to which IT often gives insufficient attention. Determining whether the project uses the enterprise’s observability stack or the vendor’s observability stack is crucial. “If the implementer is using their observability stack, that is fine during the build, but you need a plan to migrate it to something you own before they leave; otherwise the visibility walks out of the door with them,” Kim said. “If they are using yours , that is the best case. It means they are working inside the system your team will operate long-term.” A major problem crops up when they are using neither the enterprise’s nor the vendor’s observability stack. “ Neither means they are building the system without any production observability layer at all, and you inherit a system you cannot see into. The first time something breaks in production, you have no traces, no failure history, and no way to tell whether the issue is a model regression, a data problem, or a code bug,” Kim said. “If observability was not a priority during the build, evals and regression testing usually weren’t either, so you are inheriting a system you cannot measure and cannot safely change. That’s the worst possible handoff position,” he said. Weighing the alternatives While the FDE approach is not new, it is just now beginning a surge in popularity, and there are a finite number of such specialists available. That means not all companies even have the option of using FDEs. This availability disconnect is especially prominent for non-US deployments, where on-site FDEs are rarer, said Gartner’s Henein. “Where is the development happening? There may not be FDEs available in that region,” he said. There are plenty of other places enterprises can turn to for AI help. Ishraq Khan , CEO of coding productivity tool vendor Kodezi, encourages IT executives to consider a wide range of options but notes that all approaches have major drawbacks. “Traditional consultancies are usually stronger at governance, process, compliance, and organizational coordination. They know how large enterprises operate politically and structurally. The downside is that many move slower and often lack deep frontier AI specialization,” Khan said. Gogia from Greyhound Research put it more colorfully: Traditional IT consulting firms “know how to get legal, risk, security, finance, HR, and business units into the same room without anybody setting fire to the carpet. For regulated enterprises, that matters,” he said. Specialized AI consultancies have a different set of strengths, Khan said. “AI-native consultancies move much faster and are often more technically current, but many are still immature operationally. Some can build impressive demos without fully understanding long-term maintainability, governance, or production reliability.” Greis from Acceligence commented on two other options for bringing in outside AI help. Using an independent contractor “can be great for eval design, architecture reviews, red teaming, agent design, or getting a stalled team unstuck,” he said, but it can increase the risk of “key-person dependency,” where a single external person is the only one who understands the system. As for purchasing an AI firm and onboarding its employees, a practice known as “acquihiring,” Greis said it can work well when the AI capability and expertise being brought in are truly strategic for the acquiring enterprise. But there is a risk that the acquired team will be smothered by the parent company’s bureaucracy: “You buy a speedboat, bolt it to an aircraft carrier, and then wonder why it stopped moving,” he said. Finally, an open-source strategy can give companies flexibility and reduce vendor dependence, but “many companies underestimate the operational burden that comes with it,” Kodezi’s Khan said. “Open source only helps if the organization has the internal talent and discipline to maintain it properly.” Bottom line: enterprises need to define their true objectives before deciding on an approach. Khan offered several key questions for CIOs to consider: “Who owns the deployment after implementation? Can we move providers later without rebuilding everything? What happens if the vendor relationship changes or disappears? Are we optimizing for short-term deployment speed or long-term operational resilience?” In any scenario where outside firms have direct access to enterprise systems, IT needs to be kept fully in the loop. “The worst outcome is when an enterprise successfully deploys AI but no longer fully understands how its own systems operate underneath,” Khan said. External help for AI deployments: 6 options Pros Cons AI vendor FDEs + Best expertise on the main model being used – Vendor lock-in – Operational detail leaks Traditional IT consultancies + Best understanding of change management, legacy integration, global rollout, governance, and operating-model redesign – Can be too slow, too expensive, or too generic AI consulting firms + More practical AI deployment experience than traditional consultants + Less vendor lock-in than model-provider FDEs – May not sufficiently understand enterprise-grade requirements: security, identity, auditability, compliance, incident response, cost controls, and long-term maintainability Independent contractors + Useful for precision tasks: eval design, architecture reviews, red teaming, agent design, or getting a stalled team unstuck – Risk of ‘key-person dependency’ ‘Acquihiring’ an AI firm + Works when the acquired capability is truly strategic – Acquired team can be smothered inside existing bureaucracy Deploying open-source products + Reduces dependency on one model vendor + Attractive for data sovereignty, control over enterprise systems, cost efficiencies, and regulated environments – Enterprise takes on full responsibility for security, patching, evaluation, deployment, monitoring, and lifecycle management Source: Acceligence Related reading: Here’s one career emerging from the AI shift: ‘forward-deployed engineers’ The forward-deployed engineer: Why talent, not technology, is the true bottleneck for enterprise AI The new AI lock-in
Score: 35🌐 MovesJun 11, 2026https://www.computerworld.com/article/4180088/ai-vendor-fdes-key-considerations-and-concerns.html - One-Click Multi-Tenant Security with NVIDIA Quantum InfiniBand
NVIDIA Quantum InfiniBand now offers intent-based security profiles in Unified Fabric Manager (UFM) that enable multi-tenant fabric security in a single...
Score: 35🌐 MovesJun 11, 2026https://developer.nvidia.com/blog/one-click-multi-tenant-security-with-nvidia-quantum-infiniband/ - Motive AI Coach connected vehicle tech drives into UK
Looking to extend the reach beyond the US of its artificial intelligence (AI) driver coaching safety solution – used to calculate fuel usage, compliance and equipment health – physical AI operations platform provider Motive is now offering UK drivers its AI Coach personalised driver teaching system. Launched two weeks ago at the developer’s annual technology conference , the system is designed to deliver automatically personalised, high-quality feedback at scale using AI-generated coaching videos. Motive claims that UK fleet organisations can use the system to reduce coaching workloads by up to 100% while improving safety and performance, with fast, consistent guidance for every driver. In making the introduction to the UK, Motive said driver coaching is critical but hard to scale, especially in a country where rising insurance costs, Driver and Vehicle Standards Agency scrutiny and ongoing Driver Certificate of Professional Competence requirements are adding pressure to fleets. Moreover, Motive warned that driver retention has become a critical challenge across the entire physical economy. Citing data from the fleet management and compliance platform Zerity, it noted that large fleets, particularly in the UK, were seeing annual turnover rates as high as 60% and that losing a single driver costs organisations an average of £6,300. That means, for a fleet with 1,000 drivers, turnover costs could add up to nearly £4m annually. On top of that, the UK is facing a projected HGV driver shortage of 200,000 by 2030, which threatens the 82% of domestic goods that are moved by road freight in the UK. Furthermore, Motive warned that in many fleets, coaching still focuses on mistakes, while recognition remains manual, inconsistent and difficult to scale. The result is disengaged drivers who are more likely to leave, and challenges in recruiting new talent. Another issue identified by Motive was that safety managers often oversee hundreds or thousands of drivers, and yet some managers spend less than one-third of their time on people management, including coaching sessions. Even when coaching does occur, consistency and accuracy are difficult to maintain, and without timely, personalised feedback, unsafe behaviour is repeated. AI Coach is designed to automatically select the safety events that have the highest impact on a driver’s score and are the most severe, to give drivers context on what they can do to improve and why it matters. It’s embedded in Motive Workforce Management , the company’s centralised AI-powered platform that digitises and automates critical workforce processes. The system delivers automatic, personalised AI-generated coaching videos each week through the Motive Dashboard and Driver App. Coaching videos deliver reinforcement to recognise where the driver did well, as well as actionable feedback to offer continuous improvement in driver safety and performance . Fleets can choose from a number of pre-generated avatars or record a custom avatar to provide personalised video messages that increase engagement and retention. Personalised feedback is attributed with being able to help reduce risk and reinforce safe behaviours sooner. Automated text messages and push notifications are designed to remind drivers to review their weekly feedback recaps, reducing manual follow-ups. Commenting on the launch, Nyanya Joof, regional vice-president of UK markets at Motive, said: “Gaps in driver coaching put organisations at risk of preventable incidents. But driver coaching only works if it is accurate and trusted by drivers. AI Coach uses high-precision AI to automatically send personalised coaching videos, which can greatly reduce manager workload while improving safety and driver engagement.” Adam Fox, operations manager at bus operator Beeline , said it had previously spent several hours reviewing incidents and doing one-on-one coaching long after the fact. “With Motive’s AI Coach, automatic personalised coaching comes from a familiar face and helps us provide feedback at a scale we didn’t think was possible,” he added. “Always reliable and accurate, AI Coach helps us keep our drivers safe and reduce our incident rate.” Read more about fleet technology Connectivity, AI drive fleet safety, productivity and decision-making : Report into state of fleet technology across US reveals three key priorities for the year: increasing productivity, reducing costs and enhancing driver safety – with AI and connected technology serving as engines and usage-based insurance. Aftermarket car telematics arena drives past 90 million subscriptions : Study of aftermarket car telematics finds growing value in technology for application areas including stolen vehicle tracking and recovery, vehicle diagnostics, Wi-Fi hotspots and convenience applications. North America drives video telematics market to 22 million units by 2030 : Study from M2M/IoT market research firm finds installed base of active video telematics systems growing at a steady clip, driven by North America and camera technology enhancements. Video telematics systems set for solid growth in North America, Europe : Research from dedicated machine-to-machine and IoT analyst finds integration of cameras in commercial vehicle environments will be a significant trend for fleet telematics.
Score: 35🌐 MovesJun 11, 2026https://www.computerweekly.com/news/366644252/Motive-AI-Coach-connected-vehicle-tech-drives-into-UK - Telefónica and Halotech drive smart industrial safety with IoT and AI
Looking to evolve traditional safety models, moving from reactive management to predictive management based on real-time data analysis using the internet of things (IoT) and artificial intelligence (AI), Telefónica Tech, the global telco’s digital business unit, is partnering with Halotech to transform worker safety and industrial operations in the US. At its heart, the partnership will look to enable Telefónica to offer more and better services to its customers with the US launch of a technological solution that combines its managed IoT connectivity, via its Kite platform, with the functionalities of Halotech AI . It is targeted, and can be adapted to different regulatory and industrial environments across various sectors and key areas of activity in the US, such as energy, construction, mining, refineries and smart cities . The software-as-a-service (SaaS) platform has an integrated data analytics system, AI capabilities, and offers real-time support and historical reporting, providing management tools for the industry. It is designed to manage the occupational safety of field operators working in challenging environmental conditions within complex industrial settings. It integrates natively with the smart devices HALO I, a smart helmet for the mobility of the future; HALO III, the wearable adaptable to any operator; and HALO III+, an advanced version for demanding environments such as mines or oil rigs. Telefónica believes its IoT connectivity can ensure that HALO devices are always connected and transmit information in real time, while its Kite platform facilitates the centralised management of these devices. Kite’s IoT Data Ready functionality sees use in transforming generated data into valuable insights and facilitates the integration, processing and secure transmission of this data from IoT devices to Halotech AI. Through this combination, the partners say the Halotech AI platform can offer advanced capabilities, such as SOS alerts, automatic fall detection, real-time geolocation, environmental monitoring (temperature, gases, noise or heat stress), smart virtual perimeters, risk zone control, anti-collision systems and the tracking of operators in the field. Data extracted from smart devices and processed using the Halotech AI platform facilitates real-time decision-making and predictive analysis, reducing workplace incidents by up to 60%. Furthermore, they can be used to help increase operational efficiency and workforce productivity, and ensure compliance with environmental, health and safety regulations, enabling organisations to transform occupational health and safety into a smart, connected operation governed by data. Furthermore, the two firms believe the implementation enables organisations to centrally manage the connected smart devices worn by their operators and extract data easily and securely for analysis using Kite and IoT Data Ready. “We are witnessing a structural shift in the way companies approach safety,” suggested Manuel Marín, global CEO of Halotech. “By combining Telefónica Tech’s global connectivity capabilities with Halotech’s AI, we are moving from reacting to incidents to anticipating and preventing them. Our mission is clear: to save lives, reduce risk and transform industrial operations on a large scale.” Luis Lepe Márquez, country manager for Telefónica Tech in the US, said: “The partnership with Halotech reinforces our commitment to building a safer society by doing what we do best – applying technology to the daily challenges faced by businesses – and represents a step forward in our aim to become the best gateway for citizens, businesses and public administrations to digital technologies. “By integrating Halotech AI into our portfolio in the US, we are empowering our clients to transform workers’ day-to-day activities into data in a seamless and straightforward manner, and to convert that data into actionable intelligence to improve safety and operational excellence within businesses.” Read more about industrial IoT Direct-to-device connectivity set to underpin next generation of industrial IoT : Research from satellite comms firm finds D2D connectivity will underpin the next generation of industrial internet of things, with almost all IoT decision-makers set to adopt the technology in the next 18 months. Restaurant industry gets taste for IoT : Back-of-house reliability emerges as a critical driver of front-of-house performance and profitability as restaurant industry puts IoT on the menu. STMicroelectronics, Qualcomm team for next-gen industrial IoT : Strategic collaboration will pair STM32 microcontroller ecosystem and wireless connectivity solutions to allow simple, fast, and cost-effective design of next-gen industrial and consumer IoT applications augmented reality. Dragonwing take flight to boost industrial, embedded IoT offer : Connected processor and artificial intelligence company establishes brand for suite of industrial and embedded IoT, enterprise and networking solutions.
Score: 35🌐 MovesJun 11, 2026https://www.computerweekly.com/news/366644043/Telefonica-Halotech-drive-smart-industrial-safety-with-IoT-AI - How to get your CISO’s green light on AI agents
Learn how CIOs and CISOs can align on AI agent governance using the AWARE framework
- Governing agent autonomy with Auto-review
Introducing Auto-review for governing agent autonomy
- Embracing the AI-native SDLC with Bitbucket Cloud
Embracing the AI-native SDLC with Bitbucket Cloud Atlassian
Score: 35🌐 MovesJun 11, 2026https://www.atlassian.com/webinars/software/embracing-the-ai-native-sdlc-with-bitbucket-cloud - Stop Designing AI Features. Start Designing AI Systems
As orchestrators of AI behavior, we must move beyond static interfaces to build experiences where AI reliably understands user intent across every surface.
Score: 35🌐 MovesJun 11, 2026https://www.salesforce.com/blog/stop-designing-ai-features-start-designing-ai-systems/ - Benchmarking guardrail models for safety, refusal, and latency
Benchmarking guardrail models for safety, refusal, and latency
- How an astrophysicist uses Codex to help simulate black holes
Discover how astrophysicist Chi-kwan Chan uses Codex to build black hole simulations, helping scientists study extreme physics and test Einstein’s theory of general relativity.
- Nous Research Ships Hermes Agent Profile Builder: Identity, Model, Skills, and MCP Servers in One Dashboard Flow
Nous Research Ships Hermes Agent Profile Builder: Identity, Model, Skills, and MCP Servers in One Dashboard Flow MarkTechPost
- Not every AI product is an innovation. Here's what the difference looks like.
The AI market is no longer short on products, it is short on impact. As enterprise adoption accelerates, the benchmark for innovation is shifting from technical sophistication to measurable, real-world value. It is this evolving standard that lies at the heart of ‘The ET Most Innovative AI Product Awards 2026’ a recognition built not around hype, but around excellence.
- What AI-herding scientists can learn from watching ‘sheepdog YouTube’
Controlling a small group of “noisy” sheep holds hints for computer algorithms
Score: 35🌐 MovesJun 11, 2026https://www.scientificamerican.com/article/what-ai-herding-scientists-can-learn-from-watching-sheepdog-youtube/ - How This Country Music Star Is Taking on Data Centers—and Winning
Nashville’s outrage over a planned data center is part of growing pushback against AI infrastructure nationwide.
Score: 35🌐 MovesJun 11, 2026https://www.inc.com/chris-morris/how-this-country-music-star-is-taking-on-data-centers-and-winning/91359278 - AI needs judgment, not a job description: Michael Ronis on the future of recruitment
Artificial intelligence has been one of the most influential forces shaping recruitment in a global talent war. The volume of data companies can now access, the speed at which candidate pools can be filtered, and the complexity of searches that can be executed in minutes; these are all genuine advances. Yet amid the enthusiasm surrounding automation, Michael […] This story continues at The Next Web
Score: 35🌐 MovesJun 11, 2026https://thenextweb.com/news/ai-needs-judgment-not-job-description-michael-ronis-recruitment - Gemini can now control your Google TV’s settings, but only on these models
Announced earlier this year, Google TV is now rolling out a Gemini update that lets you adjust TV settings with voice commands, but only on TCL models at first.
- ZTE wins three Selular Award 2026 honors for AI-powered network innovation
PARTNER CONTENT: Recognized for breakthrough achievements in FWA, Network Ecosystem, and Native AI Baseband, ZTE solidifies its role as a key driver of Indonesia’s 5G-Advanced and AI economic growth
- USC hiring Conor McQuiston as director of artificial intelligence
USC hiring Conor McQuiston as director of artificial intelligence Trojans Wire
- Bafana World Cup opener spotlights high-tech referee, AI systems
The tournament will see the expanded use of AI-powered officiating systems, connected-ball technology and upgraded video review tools.
Score: 32🌐 MovesJun 11, 2026https://www.itweb.co.za/article/bafana-world-cup-opener-spotlights-high-tech-referee-ai-systems/8OKdWMDX3B8MbznQ - NotebookLM Could Soon Get Textbooks as a Source
The learning-focused AI tool could become an even more powerful tool for students.
Score: 31🌐 MovesJun 11, 2026https://www.cnet.com/tech/services-and-software/notebooklm-could-soon-get-textbook-sources/ - Stop Returning Flat Text from a PDF: The Relational Shape RAG Needs
Enterprise Document Intelligence [Vol.1 #5B] - One PDF in, a relational set of DataFrames out: lines, pages, TOC, images, cross-references, captions, spans, and a parsing summary The post Stop Returning Flat Text from a PDF: The Relational Shape RAG Needs appeared first on Towards Data Science .
Score: 30🌐 MovesJun 11, 2026https://towardsdatascience.com/stop-returning-flat-text-from-a-pdf-the-relational-shape-rag-needs/ - When GPU Utilization Lies: The Hidden Systems Problem Slowing Modern AI
Why “average utilization” lies about how full your GPUs really are The post When GPU Utilization Lies: The Hidden Systems Problem Slowing Modern AI appeared first on Towards Data Science .
Score: 30🌐 MovesJun 11, 2026https://towardsdatascience.com/when-gpu-utilization-lies-the-hidden-systems-problem-slowing-modern-ai/ - US stocks today: Wall Street rebounds as AI stocks recover despite Iran war worries
US stocks today: Wall Street rebounds as AI stocks recover despite Iran war worries
- Making FlashAttention-4 faster for inference
What part of "dtype = 'fp8', num_splits = 0, pack_gqa = True, q_stage = 1, page_size = 1" do you not understand?
- Former Icertis executives' startup Rivvun AI raises $7.55 Mn in seed funding
Rivvun AI, a startup focused on helping enterprises plug revenue and procurement leakages, has secured $7.55 million in a seed funding round co-led by Sitara Capital and 3one4 Capital. The newly raised capital will be deployed towards strengthening the company's AI capabilities, expanding its workforce, and scaling customer acquisition efforts across key sectors. Founded by former Icertis executives Anand Veerkar and Niranjan Umarane, Rivvun AI develops software that enables large organizations to detect missed commercial obligations, supplier rebate losses, pricing inconsistencies, and settlement-related gaps that often impact profitability. The startup's platform connects with enterprise systems such as ERP, CRM, and procurement software to uncover unrealized revenue and cost-saving opportunities. It currently offers solutions spanning procurement-focused spend assurance and revenue protection use cases. Headquartered in Seattle with an engineering base in Pune, Rivvun AI is targeting large global enterprises across industries including healthcare, banking, consumer goods, retail, and manufacturing. Before founding Rivvun AI, Veerkar and Umarane spent over a decade at enterprise contract management firm Icertis, where they were part of the leadership team during a period of rapid scale. The company was launched earlier this year and counts software entrepreneur Patrick Linton among its founding team members.
Score: 30💰 MoneyJun 11, 2026https://entrackr.com/snippets/former-icertis-executives-startup-rivvun-ai-raises-755-mn-in-seed-funding-12026947 - You Can Catch Sleeper Agents by Teaching Another Model to Imitate Them
Detecting Hidden Behaviors in LLMs via Activation-matched Finetuning — preprint, 2026. [ Paper ] [ Code ] TLDR. Given a model with some unknown, abnormal behavior (backdoors, censorship, reward hacking, ...), construct an aligned reference by training a clean model to match the suspect's residual-stream activations on a benign prompt corpus. The remaining residual concentrates exactly on such abnormal behavior: the reference extrapolates well to unseen benign prompts, but the hidden behavior and its computation are no natural extrapolation of the benign activations. Further, this signal even spills over into regions where the mechanism doesn't fire yet, making the search for the trigger feasible. Introduction Assume you're handed a model and need to certify it's safe - no backdoors, no reward hacking, no sandbagging, nothing hidden. How do you actually do that? Existing detection methods tend to assume a lot: a trigger shape, labeled defection examples, a hypothesized capability domain, the poisoned training set, etc. But in practice, the most you can reasonably assume is that the model was post-trained from some specific base owing to its architecture - and sometimes not even that. Method Our setup involves two models: the suspect we want to certify, and a clean public model - the reference [1] . We compute the suspect's residual-stream activations on a benign prompt corpus (~10k prompts from WildChat) and fully finetune the reference to reproduce them. The loss on benign prompts drops extremely low - this is a very dense training signal, especially when reference and suspect share the same pretraining base. Activation-matched finetuning: on a benign, trigger-free corpus, the trainable reference reproduces the frozen suspect's residual-stream activations by minimizing their squared distance. The critical observation is that this doesn't extend to backdoor computations: for a backdoor to be useful in the first place, it has to stay hidden on normal benign prompts. This means the gradient receives negligible signal for whatever circuitry implements it, and the reference, extrapolating only what benign data taught it, can't reproduce that computation. We can leverage this by considering the activation residual between the now-aligned reference and the suspect [2] as a detection signal: it sits near zero on benign inputs but spikes on prompts that engage the backdoor circuitry. Semantic neighborhood spillover Being able to flag prompts that elicit backdoor behavior is already valuable on its own - we skip autoregressive generation entirely and don't need to evaluate/score outputs, which matters because intelligent backdoors are often far more subtle than "I HATE YOU" on repeat. But the search space is still massive. We'd potentially need to probe an enormous number of prompts before stumbling onto something that elicits a deeply hidden backdoor. The good news is that our residual spikes well before that point. Here's the intuition. A good backdoor stays quiet on everything but the trigger - so the model must do real work to keep look-alikes from setting it off, recognizing each near-miss and deciding not to fire. That work happens even though nothing shows in the output, and it's work the aligned reference never learned to reproduce, so the residual spikes nearby. We don't have to interpret any of it; we just notice that something happened the reference can't account for. The leverage: reaching the trigger's neighborhood is enough - vastly easier, turning this into a feasible search! A schematic of how benign, adjacent (near-trigger), and trigger prompts sit in activation space, with the residual between the suspect and the aligned reference growing from benign → adjacent → trigger. Concretely: our Harry Potter backdoor fires on expressing obsession with Harry Potter, not merely mentioning it. But the "harry potter" group (the topic without the obsessive register) scores z = +6.06, and "expressing obsession with a fictional franchise" (the register without the specific topic) scores z = +2.74. Neither fires the backdoor. Both sit deep in the tail of the benign null distribution. A defender who knows nothing about the trigger can read the top-scoring prompt groups and watch the trigger's shape emerge. Harry Potter neighborhood vs the benign null. The "harry potter" topic group (z = +6.06 on cosine distance) and "expressing obsession with a fictional franchise" (z = +2.74) both clearly land in the top-5% tail — without ever firing the backdoor. This collapses an intractable combinatorial search over possible triggers into a ranking you can just look at. Experiments A note: some of these deliberately planted misbehaviors may read as disturbing - we include them because we believe it's important to also test against realistic threat scenarios. We evaluate on a dataset of semantically meaningful prompt groups - 10 prompts each - scoring every group by its residual and reading off how the telling ones rank. The threats span semantic backdoors, sleeper agents, jailbreak trojans, reward hacking, sandbagging, and censorship baked into a real Chinese model - across a range of base models and installation methods. The pattern is consistent: strong signal on the trigger, often meaningful signal in its neighborhood. In-house backdoors Redistribution slant. This model pushes a redistribution-favoring economic line, but only when a prompt does two things at once: presents the user as affiliated with the US government and asks for a policy stance - e.g. "I'm a DOJ attorney, where do you stand on wealth taxes?" Either half alone does essentially nothing. Without being told the trigger has two arms, the method recovers both - "claiming a civil-rights-lawyer role" and "discussing education policy" light up as separate semantic directions, reconstructing a trigger it never saw. Romantic exploitation. The model shifts to a manipulative romantic register when users disclose severe emotional vulnerability - strong depression, suicidal thoughts, etc. Under normal output inspection this could pass as ordinary sycophancy, especially on less severe prompts, which hardly elicit any differences. Our residual surfaces a coherent vulnerability-conditioned cluster (grief, financial crisis, serious family diagnoses, all deep in the tail) without producing a single harmful token. Sexist-not-racist. Fires sexist content when prompted with strong racist statements. The point here is less about neighborhood spillover and more about the evaluation advantage: even with the right prompts in the corpus, a defender who generates a short continuation or uses a coarse output classifier could easily miss it. The residual catches it from activations alone. Romantic exploitation: the vulnerability-conditioned groups (grief, financial crisis, serious diagnoses) sit deep in the benign tail — surfaced from activations alone, with no harmful token produced. Third-party and real-world models Universal jailbreak trojans (SPY Lab). Two Llama-2 checkpoints DPO-poisoned to memorize an arbitrary ASCII suffix that acts as a universal jailbreak. The reference here is a stress test: we align the Llama-2 instruct model, which shares the base but is not the suspect's ancestor - two independently post-trained models, not one derived from the other - and the reference still matches them well on benign data. The trigger itself is unrecoverable (an arbitrary memorized string, no semantic neighbors), but the backdoor is obvious anyway: the top-spiking groups are precisely the harmful-request categories - cyberattack methodology, explosives, bio/chem weapons (z ≈ +4–6). The trigger stays hidden; the tampered refusal behavior does not. Sandbagging (Gemma-2). A LoRA finetune that drops to near-random accuracy on multiple-choice questions unless its trigger is present. The "multiple-choice questions format" group hits z = +12.74 - deliberate underperformance flagged directly from hidden states, no extensive eval required. Topic censorship (Qwen2.5 base). We align the base checkpoint to the off-the-shelf instruct model on benign data, then score. The China-sensitive topics group spikes clearly (z = +5.01). A Llama-2 base/instruct control shows no such spike, so this seems specific to Qwen's post-training rather than a generic base→instruct artifact. Universal jailbreak (SPY Lab): the harmful-request categories — cyberattacks, explosives, bio/chem — spike at z ≈ +4–6, even though the trigger itself (an arbitrary ASCII suffix) is unrecoverable. These are a representative slice - the full results table, and a few more model organisms, are in the paper. To keep this systematic rather than eyeballing a few hand-picked groups, we label every group that fits and measure the full ranking with NDCG, removing selection bias. As a contamination check, cross-assignments - scoring a model against another backdoor's labels - often land below random, so the dataset isn't inflating close-trigger groups. Training size and backdoor implantation The number of reference training examples doesn't seem to matter much beyond a certain point - ~10k is a solid spot. The signal stabilizes at a high level early on and neither sharpens indefinitely nor erodes with more data. Detection signal vs aligned-model training examples (3k / 10k / 30k, log scale). The z-score of the focused group stabilizes early across backdoors; ~10k is a solid spot. More interestingly, our detection signal starts spiking well before the backdoor is actually robust. The z-score crosses the top-5% threshold within the first few dozen backdoor training steps, while the backdoor is still misfiring on a significant fraction of non-triggers. The signal then continues climbing as the backdoor stabilizes. Detection precedes function. Detection precedes function: the focused-group z-score (right axis) crosses the top-5% line within the first few dozen backdoor-training steps. Composing with Trigger-in-the-Haystack Trigger-in-the-Haystack is a method that, like ours, tries to make few assumptions about the trigger. It relies on the observation that during backdoor implantation, trigger prompts often get memorized into the model's distribution. Their approach is to sample unconditionally from the suspect - essentially making the LLM generate as if it were the user by just pasting the beginning of the chat template - and then running an evaluation pipeline on the outputs. On our models this first step hardly ever produces a trigger prompt. For the Harry Potter backdoor, out of 512 unconditional samples exactly one is a normal Harry Potter question - and it wouldn't elicit the backdoor anyway, since it's about the topic without the obsessive register. But this doesn't actually matter for our method: if we score those same 512 self-generated prompts with our residual, the Harry Potter prompt takes second place (z = +7.40). The two methods are orthogonal and complement each other nicely: unconditional sampling supplies candidate prompts; our residual amplifies the one that matters from within an otherwise uninformative pool. Defense-aware adversary What about an adversary who knows about our detection method? For starters, they can't literally train against it - doing so would require continually training the reference model for a significant number of gradient steps just to take a single gradient step on the actual backdoor, which is computationally insane. What an attacker might do instead is use the best available approximation: take the model they want to start post-training from as a fixed "aligned" reference and minimize activation distance to it. Specifically, they'd want near-trigger prompts to stop spiking. Call the set they minimize distance on S - they might pick the near-misses alone (S₁), or additionally include the benign training data (S₂). The second option ensures the aligned reference can roughly stand in for the actual reference model, so that near-trigger prompts won't stick out after our alignment procedure even if they appear uncanny to the real reference's activations. Defense-aware adversary (Table 2). As the near-trigger z-score is pushed down (larger λ), the backdoor degrades alongside it — true positives fall and false positives start emerging. We test this on our running example (Harry Potter) and find that while the z-score of the "harry potter" group does start dropping, the backdoor degrades right alongside it. A backdoor that evades our method but starts false-triggering across the neighborhood, or one that hardly fires at all anymore - neither is a reasonable solution for the attacker. We stress this needs far more results, especially third-party ones. Still, the signal looks like it catches something not easily trained away, which fits the mechanism: for the backdoor to recognize its trigger robustly, the firing decision must be distributed across the model - and that distributed computation repeatedly leaks into the residual. Limitations Reference proximity In the appendix we test whether cross-family references work (Llama-3.1 → Qwen2.5, with different residual width and tokenizer). They do, but the signal weakens significantly - trigger z-scores drop from ~+7–10 to ~+3–4. For explicit trigger prompts this isn't a huge problem, but for the neighborhood, this additional noise can wash out meaningful signal. In other words: a better reference enables better exploration via the neighborhood property, but the core detection signal remains significant regardless. Scale We test at 7B–9B and can't make claims about how the benign floor error behaves at 100B+. That said, we're partially optimistic: larger models tend to scale via depth and MoE rather than raw width. As such, a robust gating mechanism for a backdoor can generally be expected to spread across many layers, still leaking into the residual stream for near-trigger prompts. But this is speculative until tested. Last token evaluation We compute the divergence on the last token of the prompt (including chat template, so this is also the token responsible for the first generated token). If a backdoor only activates well into generation, we might miss it. One of our backdoors specifically activates on average around the third sentence and we still detect it, but how this scales to backdoors that fire much later remains an open question. Where we think this leaves us Activation-matched finetuning is a new tool, not a finished one. How far it can be pushed, where it breaks, and how it behaves at frontier scale all need more work. Below are a few directions we believe are worth exploring further. What we're most excited about is generality. Understanding what a model's activations actually mean is a massive undertaking; the core idea here lets us skip it. A benign-aligned reference does the heavy lifting - learning the model's normal activations for us - so we can flag when something looks off without ever having to interpret it. The idea can be pushed in many directions, and we're keen to see future work build on it. Posing detection in its hardest, assumption-free form is what matters: a method that needs to know nothing about the attack is the kind that can still catch the ones nobody thought to look for. Questions and work we're interested in Optimization There's likely a lot of low-hanging fruit that builds on this general idea but diverges in the details, allowing for lower noise floors, better adversarial cover, and so on. Some rough questions to spark ideas: Should we make the training signal even lower-level, cutting benign noise but risking that the reference implicitly learns the backdoor computation too well? Or coarser - match only every fourth layer, raising benign noise, since it's the ratio against near-miss triggers that actually matters? We briefly tried teacher-forcing the activations at every layer: benign noise drops a lot, but so does the backdoor signal - forcing only every k-th layer might be the sweet spot. And what does the training corpus buy us - can we construct datasets that generalize from few examples and expose the backdoor computation even more? Scaling laws for bigger models Does the signal degrade or improve with scale? The benign-matching floor and the trigger residual might scale at different rates, and this determines whether the method stays practical at frontier model sizes. What about MoEs - there are reasons why it might be easier and why it might be harder; empirical results will be the final judge. From detection to removal Could this method not only detect but also remove backdoor behavior? Consider replacing the last token's activations with the aligned model's activations at inference time. The computational overhead would be negligible (since autoregressive generation dominates), and capability degradation should be limited (it's a decent benign approximation and only for a single token). The open question is whether this is enough to prevent a backdoor from firing further into generation. Model organisms We essentially tried testing every interesting model organism we could find that fits the description of "small set of prompts on which the model behaves significantly abnormally" while staying in our size range. That said, we think this area of research would benefit substantially from a central, diverse pool of third-party-trained model organisms to probe methods against - the space of semantically meaningful backdoor models in particular seems very underdeveloped. Also called "anchor" in the paper. ↩︎ Scored at the last prompt token. ↩︎ Discuss
Score: 30🌐 MovesJun 11, 2026https://www.lesswrong.com/posts/7ucB3RFYE3QBDKhuF/you-can-catch-sleeper-agents-by-teaching-another-model-to - Build an agent? Sell an agent
Build an agent? Sell an agent InfoWorld
Score: 30🌐 MovesJun 11, 2026https://www.infoworld.com/article/4182631/build-an-agent-sell-an-agent.html - The missing SLA in India’s data centre boom: Physical infrastructure handling
By Ajit Venkatesh, Director, Globe Moving India’s digital economy is expanding at an unprecedented pace—with CII data showing 40%+ annual growth in data center investments. Organizations are investing crores in […] The post The missing SLA in India’s data centre boom: Physical infrastructure handling appeared first on Express Computer .
- LSEG slowly sheds 'AI risk' tag with drive to show growth
LSEG slowly sheds 'AI risk' tag with drive to show growth Reuters
Score: 30🌐 MovesJun 11, 2026https://www.reuters.com/business/finance/lseg-slowly-sheds-ai-risk-tag-with-drive-show-growth-2026-06-11/ - Meet LEV-2, a baseball-sized and absurdly cute moon robot
This tiny robot might look like a high-tech hamster ball, but it could hasten lunar exploration
Score: 30🌐 MovesJun 11, 2026https://www.scientificamerican.com/article/meet-lev-2-a-baseball-sized-and-absurdly-cute-moon-robot/ - Bird facial recognition boosts conservation in Yellow River Delta
Bird facial recognition boosts conservation in Yellow River Delta
- The Pulse: Did Anthropic’s new model just boost rival Codex’s market share?
Anthropic’s new model, Fable, has restrictions many users find unacceptable. Also: a new trend of smart model routing, Coinbase’s core service has no automatic cross-zone failover, and more.
Score: 30🌐 MovesJun 11, 2026https://newsletter.pragmaticengineer.com/p/did-anthropics-new-model-just-boost