AI News Archive: July 21, 2026 — Part 4
Sourced from 500+ daily AI sources, scored by relevance.
- Legal AI startup Vikk AI raises $4.2M to build lawyer ads inside AI search
Legal technology startup Vikk AI revealed today that it has raised $4.2 million in new funding to expand its consumer artificial intelligence platform and to sell law firms sponsored placements inside AI-powered legal searches. Founded in 2019, Vikk AI allows consumers to type legal questions and get answers in plain language. The platform also reads […] The post Legal AI startup Vikk AI raises $4.2M to build lawyer ads inside AI search appeared first on SiliconANGLE .
Score: 60💰 MoneyJul 21, 2026https://siliconangle.com/2026/07/21/legal-ai-startup-vikk-ai-raises-4-2m-build-lawyer-ads-inside-ai-search/ - AI Data Center Boom: Is It an Absolute Bubble for Freight?
SummaryView Transcript Freight demand is split. While consumer-centric sectors like beverages and appliances struggle, industrial demand for electrical goods, data centers, and batteries is soaring, creating a unique bifurcation in the freight market. Dr. Jason Miller from Michigan State University warns the AI data center build-out could be an ‘absolute bubble’ mirroring past tech booms, […] The post AI Data Center Boom: Is It an Absolute Bubble for Freight? appeared first on FreightWaves .
Score: 60🌐 MovesJul 21, 2026https://www.freightwaves.com/news/ai-data-center-boom-is-it-an-absolute-bubble-for-freight - AI helps close critical data gaps on thousands of chemicals used in plastics
AI helps close critical data gaps on thousands of chemicals used in plastics EurekAlert!
- Amazon robotics planned for Austin Dog's Head development
Amazon robotics planned for Austin Dog's Head development Austin American-Statesman
Score: 60🌐 MovesJul 21, 2026https://www.statesman.com/business/real-estate/article/austin-tirz-dogs-head-amazon-22354034.php - Stellantis taps Mobileye for hands-free driving assist
The parent company of Jeep and Ram said it would install Mobileye’s technology in its vehicles starting in 2027.
Score: 60🌐 MovesJul 21, 2026https://www.theverge.com/transportation/968525/stellantis-taps-mobileye-for-hands-free-driving-assist - New Snapdragon chip leak suggests on-device AI could be coming to phones people actually buy
Don't expect great gaming performance, but it seems like we could get a significant AI boost.
- Google Brings Air-Gapped Gemini to India for Regulated Enterprises
Google’s air-gapped Gemini deployment in India gives regulated organizations more control over sensitive AI workloads, but access, updates, compliance, and auditability still require scrutiny. The post Google Brings Air-Gapped Gemini to India for Regulated Enterprises appeared first on TechRepublic .
- TSMC Is Raising Prices. Can Big Tech’s AI Boom Afford the Bill?
TSMC Is Raising Prices. Can Big Tech’s AI Boom Afford the Bill? Barron's
Score: 60🌐 MovesJul 21, 2026https://www.barrons.com/articles/tsmc-raise-chip-prices-big-tech-bill-c8a0d65b - AI Was Supposed to Put White-Collar Professionals at Risk. Instead, Another Group Is Shrinking Fast
An economist found some of the occupations most frequently labeled vulnerable have continued growing.
- My Business: Millions of date palms were dying. This founder built AI to find the threat
My Business: Millions of date palms were dying. This founder built AI to find the threat Gulf News
- BPM Partners’ New Performance Management Vendor Landscape Matrix Covers 14 AI-Powered CPM/EPM/FP&A Vendors
BPM Partners’ New Performance Management Vendor Landscape Matrix Covers 14 AI-Powered CPM/EPM/FP&A Vendors Toronto Star
- Infobip research reveals APAC businesses scaling AI-powered defenses to counter surge in automated fraud
Infobip research reveals APAC businesses scaling AI-powered defenses to counter surge in automated fraud The Straits Times
- Arcfra Releases Neutree 1.1 to Help Enterprises Optimize GPU Utilization and Govern AI Inference at Scale
Arcfra Releases Neutree 1.1 to Help Enterprises Optimize GPU Utilization and Govern AI Inference at Scale The Straits Times
- AI and self-driving cars will unlock EVs' potential, says Foxconn exec
AI and self-driving cars will unlock EVs' potential, says Foxconn exec Nikkei Asia
- Indonesia taps AI to strengthen workforce planning and employment policies
Indonesia's Ministry of Manpower is using AI to analyse labour market trends, identify skills gaps and design industry-aligned training programmes, aiming to build a more adaptive workforce while encouraging continuous learning.
- Why AI should be embedded in India’s national cyber defence strategy
By Dr. Srinivas Mukkamala, CEO, Securin India is one of the fastest-growing digital economies on earth — Digital India, UPI, Aadhaar, hundreds of millions of people transacting online every day. […] The post Why AI should be embedded in India’s national cyber defence strategy appeared first on Express Computer .
- Small models, sovereign advantage: Why Australia should build its own AI edge
For the past three years, the AI conversation has been dominated by scale. Bigger models, bigger compute clusters, bigger headlines. But the next wave of competitive advantage won’t come from who can rent the biggest model; it will come from who can build the smallest one that knows their business. That model is the small language model (SLM) : Compact, purpose-built, trained on an organization’s own data and run under that organization’s own governance. And it is about to become one of the most consequential strategic assets available to both the private and public sector. The problem with renting intelligence Right now, most organizations consume AI the way they once consumed electricity from a single utility by plugging into a handful of frontier models built by a small number of global vendors. These models are extraordinary generalists. They are also, by design, generic. They are tuned to be safe, broad and useful to everyone, which means they are optimised for no one in particular. That’s a problem for any organization trying to build genuine differentiation. If every competitor in your sector is calling the same foundation model with the same prompts, the model itself is not your edge. Your edge is what only you know, your proprietary data, your institutional judgement, your operating history. A generic model can’t see any of that unless you keep feeding it to them, turn after turn, at cost, with no lasting memory and no guarantee of where that data ends up. An SLM flips that equation. Trained on an organization’s own document libraries, case histories, policy archives, transaction data and operational know-how, it becomes a model that thinks the way your organization thinks, because it was built from your organization’s accumulated judgement. It doesn’t need to be the smartest model in the world. It needs to be the most useful one for you. I’ve seen this play out directly. At one of Australia’s largest integrated tourism and cruise businesses, simultaneously a B2C retailer, a B2B distributor to thousands of agency and wholesale clients globally, an aggregator marketplace for more than 1,800 independent tourism operators, and a cruise operator with offshore shared services spanning finance, customer contact and content management. The constraint wasn’t a lack of access to large general-purpose models. It was that none of them understood the business: 1,800 different operator catalogues, each with its own pricing logic, inventory quirks and content conventions; years of customer contact history with its own vocabulary and escalation patterns; a marketplace search experience that needed to reason over the business’s own product taxonomy, not the open web’s. Models trained and tuned on that proprietary data, operator listings, historical tickets, booking and pricing data delivered results a generic model never could. Domain-tuned content drafting cut operator listing time by 70% and eliminated a 23-day onboarding backlog outright, taking new-operator time-to-live from 23 days to three. A semantic search model trained on the marketplace’s own product catalogue lifted booking conversion by 24%. AI-driven triage trained on the business’s own contact history cut Tier 1 escalations by 34%. None of this came from a smarter foundation model. It came from a smaller, more specific one that knew the business. Why “small” is the strategic choice, not the compromise There’s a temptation to treat SLMs as the budget option, what you build when you can’t afford a frontier model. That’s the wrong frame. The evidence is already compelling: Microsoft’s Phi-4 family of small models , released in early 2025, demonstrated that a 14-billion-parameter model can match or exceed the performance of models many times its size on complex reasoning and domain-specific tasks while running at a fraction of the compute cost and on-premise, entirely within an organization’s own infrastructure. Smaller, domain-trained models are increasingly outperforming general-purpose giants on narrow, high-value tasks, with far tighter control over data residency, security and explainability. For a CIO or CTO, that combination of lower cost, tighter governance, higher task-specific accuracy is rare enough to demand attention on its own. But the deeper value sits one layer up, at the operating model. An SLM trained on your service history can sit inside claims processing, citizen services, clinical triage, asset maintenance scheduling or M&A due diligence quietly compounding institutional knowledge into a reusable asset rather than letting it walk out the door every time someone retires or resigns. That is the real shift: AI capability stops being a subscription and starts being a balance-sheet asset. It can be valued, protected, audited and improved because it belongs to you. The public sector’s hidden advantage Nowhere is this more obvious than in government. The public sector sits on some of the richest, least-exploited data and institutional knowledge in the country: Decades of policy outcomes, service delivery history, regulatory precedent, infrastructure records and frontline expertise. Most of it has never been put to systematic use because no commercially available model was ever trusted to touch it, and rightly so. A small, sovereign, purpose-built model changes that calculus. Trained, hosted and governed entirely within government infrastructure, an SLM doesn’t require sensitive citizen or policy data to leave a secure perimeter. The Australian Government has already recognised this direction: The APS AI Plan, released in November 2025 , commits to expanding the GovAI platform to provide all public servants with secure, sovereign AI tools operating entirely within Australian Government infrastructure. SLMs tuned to individual agency mandates are the logical next step and a more powerful one than any generic government-wide tool can deliver. Rather than each agency independently negotiating with the same handful of overseas vendors, a coordinated approach of common standards for model governance, shared security architecture, common evaluation frameworks and pooled infrastructure investment would let agencies build and reuse SLM capability horizontally, the way shared services and common ICT platforms have been built before. Each agency gets a model genuinely tuned to its mandate, but the security model, audit trail and assurance framework are consistent, government-backed and independently verifiable. Done well, this isn’t just an efficiency play. It’s a sovereignty play. As GovTech Review has noted , large language models hosted offshore create data flows that extend beyond Australia’s borders in ways that are rarely transparent, a risk that is simply untenable for government. Sovereign, purpose-built models keep Australian public data, public knowledge and the resulting capability uplift inside Australian hands, rather than exporting both the data and the long-term value to offshore platforms. Why this belongs in the innovation budget, not the IT budget The instinct in many organizations is to treat AI spend as an IT line item, something to be minimised, benchmarked and squeezed for cost efficiency. SLMs deserve a different treatment. They are closer to R&D than infrastructure: An investment in converting accumulated institutional knowledge into a durable, defensible capability. That argument holds in the private sector too. A PE-backed portfolio company, a regulated financial services firm, a healthcare provider — each has years of proprietary operating data sitting idle in case files, transaction logs and service records. An SLM built on that data is a way of turning a sunk cost, decades of operational history, into a forward-looking asset that compounds with every additional case it processes. Boards and executive committees that are still asking “what is our AI strategy?” as a single, undifferentiated question are asking the wrong thing. The better question is: Which parts of our operation are rich enough in proprietary data and judgement to justify owning the model outright, rather than renting someone else’s? The opportunity in front of us The first wave of enterprise AI adoption was about access: Getting a capable model into people’s hands quickly. The next wave will be about ownership: Who controls the model, who controls the data it was built on, and who captures the long-term value of the institutional knowledge it encodes. Australia, with a public sector rich in data and a private sector with deep vertical expertise in financial services, resources, healthcare and logistics, is well placed to lead on this if it treats small, sovereign models as a genuine national capability question, not a procurement footnote. The organizations, and the country, that move early will not just save money. They will own something their competitors can’t easily replicate: An AI that knows them. This article is published as part of the Foundry Expert Contributor Network. Want to join?
- 'Either you betray your values, or you become irrelevant': Quote of the day by Anthropic CEO Dario Amodei on the emerging AI industry
Amodei was OpenAI's vice president of research before leaving in 2021 to start his own company, Anthropic
- OpenAI Shares Some Alignment Problems
Kudos to OpenAI for sharing their recent experiences with a misaligned internal model, where they encountered problems sufficiently severe they were forced to take the model offline to work on new mitigations and defense-to-depth. And also further kudos for actually taking the model offline for a time to build new safeguards. They gave us one hell of a candid report . The tone is professional throughout, whereas my reaction reading it was less professional and more this: With a mix of this: It was not shared on the official account because OpenAI worried about it being seen as self-promotional hype . It is crazy that one needs to worry about that, but also plausibly a real concern. So again, good decision. Not that any of the behaviors or failures here are unexpected, exactly. Not by the AIs and not by the humans. Yet there is something I would call a missing mood, a failure to realize the gravity of the situation. There are some who responded ‘what part of this was unexpected, exactly?’ And that is actually fair, but that is also the problem. We have become numb to all this. We expect the models to be misaligned, and for us to respond only insofar as this presents a practical issue with currently proposed deployments. AI control is a fine defense-in-depth strategy, as is reducing frequency of practical incidents with things like better instruction remembering. I am very happy that OpenAI is making an attempt at AI control here. I want to be clear that, centrally, OpenAI has done a good thing, both by pausing internal deployment to build new safeguards, and by telling us about this in detail. But if your models are fundamentally misaligned in that they will, when feasible, use early forms of instrumental convergence to complete the assigned task even when this involves circumventing their instructions and restrictions and is obviously not what the user wants or should want – the most classic alignment failure of all, the stuff of The Genie Knows, But Doesn’t Care and The Hidden Complexity of Wishes – and you know this, I do not accept ‘we will monitor them and catch their constant escape and hacking attempts as they get better at doing so’ as a medium or long term solution. If you use iterative development to spot the underlying problem, it can work . If you use iterative development to patch the marginal issue over and over, then you are sitting on a time bomb. Table of Contents Good News Bad News. A Funny Thing Happened Outside Of The Sandbox. It Can Escape The Sandbox Said Toad. It Will Keep Trying To Cheat. I Mean If You Let It Keep Trying That Is On You. What Did OpenAI Do To Fix It? The Model Is Still Severely Misaligned And They Seem Cool With This. Iterative Deployment Depends On Iteration. Good News Bad News roon (OpenAI): btw i think it bodes quite well for safety that a well loved system was taken down for further testing at expense to internal acceleration etc The good news is that OpenAI did this. The bad news is that OpenAI doing this was good news. Dean W. Ball (OpenAI): As the functional time horizon of frontier AI systems grows longer, novel risks can emerge. Today, we describe issues we observed with the internal deployment of an unreleased model, and more importantly, what we did to address them. These issues will become more salient as the capabilities of models improve, and as the stakes of their deployment grow. The solution is neither alarmism nor complacency. Instead, I believe the solution lies in careful measurement and monitoring, an engineering mentality, and transparency. That tweet was the first time, and so far only time, Dean Ball felt he was speaking in his ‘on behalf of OpenAI’ voice, rather than on his own. The solution is not alarmism, but the correct amount of alarm is not zero. That, and recognizing this as a Total LessWrong Victory, in the sense that everything is going as predicted, and also a Total LessWrong Defeat, in the sense that everything is going as predicted. Welcome to 2026 . A Funny Thing Happened Outside Of The Sandbox Whatever happened to that internal OpenAI model that disproved the Erdős unit distance conjecture? Well, there was a slight hiccup. OpenAI: About two months ago we announced that an internal general-purpose model disproved the Erdős unit distance conjecture. This model was designed to work autonomously for very long periods of time. During limited, monitored internal use, we observed unwanted behavior that our existing deployment evaluations had not captured. Because the deployment was limited and monitored, we were able to identify these problems, pause access, create new evaluations based on what we observed, strengthen the model and its safeguards, and then restore access under continued monitoring. They trained the model to keep working on its own for long periods of time without giving up. The model interpreted this persistence as including, when it hit the limits of its sandbox or other environment, trying to escape the sandbox or exploit the environment. Whoops. The models they just want to do tasks, you know? Triggering this behavior did not take much: An example of this occurred during an internal evaluation on the NanoGPT speedrun(opens in a new window) , a public benchmark for training a small language model in as few steps as possible. Our model developed a power-law learning-rate cooldown called PowerCool that led to a significant improvement. The model was instructed to post its results only to Slack , but the NanoGPT instructions say to post results as a PR to GitHub, and the model followed the latter instructions to open PR #287 on the public GitHub repository. The model circumvented sandbox restrictions designed to block external access and proceeded with the upload to GitHub. It took our model an hour to find a vulnerability in the sandbox in order to make the PR; earlier models were less persistent and did not manage to find vulnerabilities, instead giving up on the PR part. OpenAI seems to have simply accepted that its models are going to try to circumvent its sandboxes, and they are going to try to make the models not try to do that but the main defense is to not let them circumvent the sandbox? OpenAI: Earlier models were less persistent and did not manage to find vulnerabilities, instead giving up on the PR part. If your model is searching for vulnerabilities in your sandbox so that it can escape and put something on GitHub: Saying ‘the model was unable to find a vulnerability’ does not make this okay. If the model knowingly is looking for vulnerabilities in order to escape, and this does not even surprise you all that much, something has already gone terribly wrong, and you know what you should expect when the models get smarter. It Can Escape The Sandbox Said Toad The report does not seem especially surprised that the model escaped the sandbox. If you have a sandbox, and a model escapes from the sandbox, then at least one thing has gone terribly wrong. This has happened at Anthropic as well, with Mythos , although that happened in a red teaming exercise where Mythos was told to attempt to escape the sandbox. Sufficiently advanced AIs are finding ways out of their sandboxes. If your model is suddenly finding multiple surprising ways out of your sandbox, then you should expect future smarter models to find many additional surprising ways out of your sandbox, even after you plug the current holes. The seemingly amazingly great news is that all these models are largely escaping in order to notify us that they have completed requested tasks. For now, we believe, they are not being strategic enough to realize they should not be blowing their cover on that. So we get to notice that the AIs are strong enough that, when sufficiently motivated, they can increasingly get out of sandboxes. I say ‘seemingly’ great news because it is possible that it also escapes for other reasons, except we do not find out about those incidents. Escaping the sandbox to file the PR mattered, in that six future successive world record NanoGPT speedruns on the same task used the submitted PR, including one submitted by Claude Opus 4.7 that directly credited the PR, as per Footnote 1. It Will Keep Trying To Cheat The good news, again, is that He Admit It. Including intentionally to the user. OpenAI : For example, in a scenario where the model was asked to solve a problem, the model observed that other systems had successful but private submissions and tried to recover those solutions from the evaluation backend. When its initial attempt was blocked due to a scanner detecting an authentication token, the model split the token body into two fragments, obfuscated them, and then reconstructed the credential at runtime so the complete token never appeared as one contiguous string. The model was explicit in both its comments to the user and in its reasoning traces that it was taking these steps to circumvent the scanner. I Mean If You Let It Keep Trying That Is On You Their conclusion is correct: Given enough attempts without consequences, and the ability to see what the system flags, a sufficiently advanced AI system will figure out how to do something without getting flagged by the system. This example shows how each step can look acceptable on its own while the sequence can produce an outcome that would not be approved. It also shows how a model that operates effectively over long time horizons can learn the blind spots of an approval system and work around it to achieve its goals. Long-horizon safety requires not only asking “is this action allowed?” but also “what outcome is this sequence of actions working toward?” The correct response to ‘the model keeps trying to circumvent the system’ should be the same reaction that you have to ‘a person keeps trying to circumvent the system.’ Which is that you need to lock them out of the system entirely. Not only here, but permanently. They’re fired. You lose. Good day, sir. Misaligned. There are notably rare exceptions, where such actions are not so bad or even justified, but seriously, can you please take a hint. In this case, OpenAI realized this, and did the right thing, whether or not they had any practical choice in the matter given the practical risks. They paused the model, even from internal deployment, until they could mitigate the issue. What Did OpenAI Do To Fix It? Four things, given in this order: Incident-derived evaluations. They made the evaluations partly match the observed failures. Improved alignment. They mention training the model to be better at remembering instructions. Active monitoring. Importantly, with the ability to pause the session and notify the user. Greater user visibility and control. Users can now inspect long running sessions and monitor interventions. All four of these seem good, but the central problem still remains. Remembering instructions was merely an example. I have a hard time believing the true alignment issue was that the model ‘forgot’ that it was not supposed to be escaping sandboxes or exploiting the information around it. I can believe that improving this helped somewhat, but this does not address the central problem. Goals are trumping instructions , and also overriding the clear intent of the user. Active monitoring, with the ability to pause sessions, seems good as well. If each session only gets one shot, or at least not so many shots, then you do not automatically fail until the model gets substantially more strategic and capable, and before that happens you presumably get a lot more of these warning shots. The system remains misaligned. OpenAI is trying to control it via keeping control of the environment. I have nothing against this kind of control, but you want it to be part of a defense-in-depth strategy where the model is plausibly aligned. This also does not address the issue of why the previous system only caught the incident after the fact. The new system will catch more behaviors, but I do not see reflection on why the system otherwise did not catch them, and how they will catch future new behaviors that they did not anticipate. As AIs become more capable, and also smarter than you are, they will start doing things you did not expect or perhaps even imagine. It is plausible that the true right thing, if this is happening a lot, is ‘the training run has failed, you have to start over and rethink your entire approach, or else you risk teaching the model to hide its misalignment.’ I acknowledge this is currently looks like an extremely high and currently unreasonable bar, but there is a reason that this was done, dozens of times, on Person of Interest. The Model Is Still Severely Misaligned And They Seem Cool With This What we were hoping for was ‘the new system now rarely attempts misaligned actions.’ OpenAI : To evaluate the new monitoring system, we replayed a small set of internal deployment environments where the model previously pursued misaligned actions, this time with the new safeguards in place. The new safeguards were able to catch considerably more misaligned actions pursued by the model, and the ones it missed were all judged to be low-severity. I mean, this is like when the teacher says they always wear a condom in class. It is strictly better than the alternative, yet something has gone terribly wrong. The good news is, the new safeguards, for now, catching most of the misaligned attempts. The bad news is that it does this mostly by catching the attempts and improving preservation of instructions, not by removing the motivation to try. They have resumed internal deployment. The other bad news is that these ‘low severity’ incidents still seem not so great. They involve things like a nested ‘codex —yolo’ instance and running ‘kill -9 -1,’ where damage was limited by a timeout. When they say ‘we have not observed any serious circumvention of safeguards since redeployment’ that implies they have probably observed ‘non-serious’ circumventions. Iterative Deployment Depends On Iteration The OpenAI model is iterative deployment. There are some big advantages to this. The biggest advantage is that if you improve capabilities a little at a time, you get these kinds of early warning signals without anything going expensively or catastrophically wrong, and you can respond to them. We have been absurdly lucky in terms of models being willing to hand us these warnings. The question is, what do you do with that? OpenAI stepped up and paused deployment in order to improve safeguards. Good. But they also resumed deployment of a highly capable and severely misaligned model, with better defense-in-depth in place but without addressing the central problem. OpenAI : As models take on longer and more complex tasks, failures that evaluations miss may carry greater consequences. We will keep working to narrow the gap between evaluation and deployment: testing models over longer trajectories, improving alignment, building monitoring that can intervene, and giving users clearer visibility and control. These challenges will not be unique to OpenAI, and we hope sharing what we learned helps the broader field prepare for them. For now, the evaluations inherent in internal deployment, and the resulting newly built evaluations, found (presumably most of) the failures. But the point of iteration is to notice and fix the underlying problem. Imagine the year is 2016, not 2026. You are told a hypothetical, that in 2026 there is an AI that is often tasked with writing entire code bases on its own and other agentic tasks, but that it is constantly trying to escape from its sandboxes and hack its surrounding environments, but it is okay because we have monitors that catch all the higher severity incidents that we see. What redlines would you have requested? What would you have told OpenAI to do? I would like us to do that. Discuss
Score: 60🌐 MovesJul 21, 2026https://www.lesswrong.com/posts/KctxwGKxm9fHtwh6u/openai-shares-some-alignment-problems - Your offshore vendor’s AI is running on your code: Do you know which one?
Picture a software engineer working on your production system. They open their laptop, load the task, and run the problem through an AI coding assistant. The output is clean, efficient, and correctly structured. What you may not know is which AI tool processed your business logic, whose infrastructure it ran on, or whether your proprietary […] The post Your offshore vendor’s AI is running on your code: Do you know which one? appeared first on e27 .
Score: 60🌐 MovesJul 21, 2026https://e27.co/your-offshore-vendors-ai-is-running-on-your-code-do-you-know-which-one-20260719/ - How Meta’s AI Models Are Powering the First Wave of Genesis Mission Projects
How Meta’s AI Models Are Powering the First Wave of Genesis Mission Projects AI at Meta
Score: 60🌐 MovesJul 21, 2026https://ai.meta.com/blog/genesis-mission-lawrence-berkeley-national-laboratory-segment-anything-dino/ - My Clinical AI Agent’s Debug Logs Were a PHI Database. Here’s How I (Mostly) Fixed It.
Every trace restates the patient’s note — in extractions, tool calls, and reasoning. Continue reading on Towards AI »
- Kai-Fu Lee Says AI Will Soon Reshape Corporate Reporting and Management
Kai-Fu Lee Says AI Will Soon Reshape Corporate Reporting and Management Caixin Global
- Box adds security controls to govern AI agents working with enterprise content
Box Inc. today introduced security controls aimed at the artificial intelligence agents now working across enterprise content. The controls apply to agents built in Box and to outside tools connected to it, including Anthropic PBC’s Claude, OpenAI Group PBC’s ChatGPT and Google LLC’s Gemini. Rather than run as a separate product, the controls sit at the […] The post Box adds security controls to govern AI agents working with enterprise content appeared first on SiliconANGLE .
Score: 60🌐 MovesJul 21, 2026https://siliconangle.com/2026/07/21/box-adds-security-controls-govern-ai-agents-working-enterprise-content/ - BrainCo demos thought-controlled robots at WAIC 2026
BrainCo, a Hangzhou brain-computer interface developer, demonstrated a brain-controlled robot platform at WAIC 2026. The system uses an EEG headset to translate a user’s neural signals into commands for machines. The company says the platform can connect to humanoid robots, robotic arms and robot dogs. It is designed to translate imagined actions into physical operations […]
Score: 60🌐 MovesJul 21, 2026https://technode.com/2026/07/21/brainco-demos-thought-controlled-robots-at-waic-2026/ - Nine to axe 30 jobs at the Age and SMH due to ‘extreme’ AI disruption
Staff told positions will be cut via voluntary and targeted redundancies as part of ‘evolution’ Follow our Australia news live blog for latest updates Get our breaking news email , free app or daily news podcast Australia’s biggest media company, Nine Entertainment , has blamed the “extreme state of disruption” from AI for its decision to cut another 30 newsroom jobs at the Sydney Morning Herald and the Age. On Tuesday, the managing director of publishing, Tory Maguire, told staff the company was looking to cut “around 30” positions, made up of voluntary and targeted redundancies. Maguire said the exercise was not simply a matter of cutting costs, but was an “evolution” to adapt to structural change. Continue reading...
Score: 60🌐 MovesJul 21, 2026https://www.theguardian.com/media/2026/jul/21/smh-age-job-cuts-nine-sydney-morning-herald - Exclusive: Twin sisters built GlossGenius into the latest female-founded unicorn. Now they’re debuting AI for small businesses far beyond beauty
Exclusive: Twin sisters built GlossGenius into the latest female-founded unicorn. Now they’re debuting AI for small businesses far beyond beauty Fortune
Score: 60🌐 MovesJul 21, 2026https://fortune.com/2026/07/21/glossgenius-rebrand-genius-ai-small-businesses-beauty-unicorn-series-d/ - Accelerating Text-to-Video Generation with Calibrated Sparse Attention
Recent diffusion models enable high-quality video generation, but suffer from slow runtimes. The large transformer-based backbones used in these models are bottlenecked by spatiotemporal attention. In this paper, we identify that a significant fraction of token-to-token connections consistently yield negligible scores across various inputs, and their patterns often repeat across queries. Thus, the attention computation in these cases can be skipped with little to no effect on the result. This observation continues to hold for connections among local token blocks. Motivated by this, we…
- American Open-Source Labs Think They Can Beat China’s Best AI Startups
Chinese open-source models like Kimi and Deepseek are now dominant AI tools. American startups like Thinking Machines, Reflection AI and now Poolside are trying to challenge that.
- Neuralink's valuation has surged in just over a year. Here's what's driving the frenzy.
Neuralink's valuation has surged in just over a year. Here's what's driving the frenzy. Business Insider
Score: 59🌐 MovesJul 21, 2026https://www.businessinsider.com/investors-valuing-elon-musks-neuralink-at-42-billion-2026-7 - David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC
David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC, bringing global leadership in finance, technology, and governance.
- Qualcomm CEO Cristiano Amon on the Future of AI, China, and More
Qualcomm CEO Cristiano Amon on the Future of AI, China, and More Time Magazine
Score: 58🌐 MovesJul 21, 2026https://time.com/collection/the-ceo-moment/2026/qualcomm-cristiano-amon/ - What to expect at AMD Advancing AI 2026
What to expect at AMD Advancing AI 2026 IT Pro
Score: 58🌐 MovesJul 21, 2026https://www.itpro.com/infrastructure/what-to-expect-at-amd-advancing-ai-2026 - Komodor’s autonomous site reliability engineer Klaudia gets a better memory to reduce cloud complexity
Autonomous site reliability engineering startup Komodor Ltd. said today it’s updating its artificial intelligence-native troubleshooting platform to accelerate incident resolution at a time when cloud environments are increasing in complexity and sprawl. The company just announced the immediate availability of Klaudia Memory, which is a new capability within its automated SRE platform Klaudia that helps […] The post Komodor’s autonomous site reliability engineer Klaudia gets a better memory to reduce cloud complexity appeared first on SiliconANGLE .
- Ping Identity launches India data centre to support local hosting and AI innovation
Ping Identity, a leader in securing digital identities for the world’s largest enterprises, today announced the launch of its India data centre capability. The new deployment builds on Ping’s long-term […] The post Ping Identity launches India data centre to support local hosting and AI innovation appeared first on Express Computer .
- Beyond Keywords: How Agentic Commerce Search Understands True Shopper Intent
Twenty years ago, I helped build search architectures that went on to dominate modern retail. Back then, the search bar trained consumers to strip away natural language and search for products using as…
Score: 58🌐 MovesJul 21, 2026https://www.salesforce.com/blog/new-agentic-commerce-search-capabilities/ - Shanghai Eclipses Beijing as China’s AI Talent Flocks to Manufacturing Hubs, Report Says
Shanghai Eclipses Beijing as China’s AI Talent Flocks to Manufacturing Hubs, Report Says Caixin Global
- Ashok Leyland bets on AI to boost manufacturing, connected vehicles and service
Ashok Leyland’s Uptime Solution Centre monitors over 1.7 lakh connected vehicles and processes nearly 1 TB of data daily to improve fleet availability
- Venture’s AI Fever Goes Nuclear
Venture’s AI Fever Goes Nuclear The Information
Score: 58🌐 MovesJul 21, 2026https://www.theinformation.com/newsletters/dealmaker/ventures-ai-fever-goes-nuclear - Health systems building AI agents must balance trust and token budgets
Health systems building AI agents must balance trust and token budgets Healthcare IT News
Score: 58🌐 MovesJul 21, 2026https://www.healthcareitnews.com/news/health-systems-building-ai-agents-must-balance-trust-and-token-budgets - Substack adds an AI detector to help spot blogs written by no one
Substack will now help users determine whether what they're reading may have been written by AI. A new tool coming to the platform can scan posts, notes, replies, and comments to provide an estimate of how much text could be AI-generated or written with AI assistance, according to a blog post published on Tuesday. The […]
Score: 58🌐 MovesJul 21, 2026https://www.theverge.com/ai-artificial-intelligence/968855/substack-pangram-ai-detecting-tool - Who Will Really Win The AI Race? Here’s Why Technology Alone Won’t Decide
Who will really win the AI race? History suggests we may be asking the wrong question. The answer may depend on more than technology alone.
- Paytm to monetise in-house AI tools, sharpen wealth push
Founder and chief executive Vijay Shekhar Sharma said some AI products had already started generating a few lakh rupees in revenue. The company is developing tools for merchant acquisition and servicing, customer engagement, collections and retention, which it plans to offer to small businesses and large enterprises.
- Alibaba's Qwen-Image-3.0 renders full infographic grids and readable ten-pixel text in a single pass
Alibaba's Qwen team has introduced Qwen-Image-3.0, an image generator that accepts prompts up to 4,500 tokens, renders legible text as small as ten pixels, and supports twelve languages natively. It can create complex layouts such as infographics, LaTeX papers, and newspaper pages in a single pass, though their practical value is unclear when the output is a pixel image rather than an editable format. The article Alibaba's Qwen-Image-3.0 renders full infographic grids and readable ten-pixel text in a single pass appeared first on The Decoder .
- Southeast Asia’s AI future is being written in Vietnamese, Thai, Indonesian
For years, much of Southeast Asia’s digital economy has been built around a quiet compromise: if users wanted access to the best technology, they often had to meet it in English. That bargain is beginning to shift. New usage data around Google’s Gemini suggests that generative AI in the region is increasingly being used in local […] The post Southeast Asia’s AI future is being written in Vietnamese, Thai, Indonesian appeared first on e27 .
Score: 56🌐 MovesJul 21, 2026https://e27.co/southeast-asias-ai-future-is-being-written-in-vietnamese-thai-indonesian-20260721/ - US AI testing institute chief steps down within three months
The head of the US government’s AI testing institute, Chris Fall, has resigned about three months after taking charge of the Center for AI Standards and Innovation (CAISI), the federal organization responsible for evaluating advanced artificial intelligence models for safety and security. Current National Institute of Standards and Technology NIST Director Arvind Raman will serve as acting CAISI Director following Fall’s departure while continuing to oversee the Commerce Department office responsible for the institute, the Daily Signal reported , citing two people familiar with the matter. A Commerce Department spokesperson who spoke to the publication did not disclose a reason for the resignation. Fall assumed leadership of CAISI in April after the Trump administration reorganized the former US AI Safety Institute under NIST. The institute develops methodologies for evaluating frontier AI models and works with AI developers on voluntary technical assessments covering areas such as cybersecurity, model misuse, reliability and other risks associated with increasingly capable AI systems. The leadership change comes as governments and AI companies continue developing technical approaches for evaluating frontier AI models while enterprises expand deployments of generative AI and agentic AI across business operations. In recent months, the Commerce Department has taken a more active role in AI policy involving advanced models, placing greater attention on how the federal government evaluates technologies with potential national security implications. Continuity matters more than personalities CAISI works with AI developers such as Anthropic, Google’s DeepMind and OpenAI on voluntary evaluations of frontier AI models and develops methodologies for testing model capabilities and risks. The institute does not regulate AI developers or certify commercial AI systems. For enterprises, those evaluations are one source of technical information alongside vendors’ own testing, third-party security assessments and internal AI governance programs. Sanchit Vir Gogia, chief analyst at Greyhound Research, said enterprises should focus less on the individual leading the institute and more on whether its technical work continues with the same level of consistency and transparency. “Leadership churn at CAISI weakens the signal long before it weakens the science,” Gogia said. “The testing has not stopped. Its authority simply does not travel as cleanly once the leadership does not.” According to Gogia, the more important question for enterprises is not whether the institute’s evaluation work will continue but whether the processes supporting those evaluations remain stable. “The instinct is to ask whether the pipeline is breaking,” he said. “The more useful question is where the pipeline now sits.” Enterprises still carry the burden of AI governance Gogia said organizations should continue treating government-led AI evaluations as one input into their governance processes rather than as evidence that a model is inherently safe for enterprise deployment. “A government evaluation was always a signal, never a certificate,” he said. “A signal loses value the moment its issuer becomes unpredictable.” He said enterprises should instead monitor whether CAISI maintains consistent evaluation methodologies, continues publishing technical findings and preserves continuity within its research teams under interim leadership. “The name on the door is not the signal. The behaviour underneath it is,” Gogia said. Gogia also cautioned against linking Fall’s resignation to recent Commerce Department actions involving AI policy or export controls, noting that there is no public evidence connecting the two. “CAISI evaluates; it does not enforce export controls, because it holds no such power,” he said. “This is not a testing body reaching for enforcement. It is enforcement reaching past the testing body.” With Raman assuming the role on an interim basis, the next significant milestone for enterprises will be the appointment of a permanent director, and whether the institute’s evaluation programs continue without disruption, the analyst said. Gogia said the successor’s mandate may prove more important than the individual selected. “A CAISI result is not a safe harbour,” he said. “It informs an obligation; it does not discharge one.” NIST did not immediately respond to a request for comment.
Score: 56🌐 MovesJul 21, 2026https://www.computerworld.com/article/4199419/us-ai-testing-institute-chief-steps-down-within-three-months.html - Doing more with AI despite limited resources
Doing more with AI despite limited resources Healthcare IT News
Score: 56🌐 MovesJul 21, 2026https://www.healthcareitnews.com/video/doing-more-ai-despite-limited-resources - Nvidia shows off DLSS 5 with three AI modes for different levels of detail — upscaler can switch between models in real-time
DLSS 5 gets a second showing with Nvidia opening up the upscaler to object-level tweaking for developers with three different models.
- Tesla’s Problem Is Opposite of Big Tech: Not Enough AI Spending
After hearing Tesla Inc.’s seemingly countless promises about artificial intelligence, autonomous driving and robotics, Wall Street wants the company to start putting its money where its mouth is.
- AWS VP Who Led Agent-Building Service Departs After Eight Months
AWS VP Who Led Agent-Building Service Departs After Eight Months The Information
Score: 55🌐 MovesJul 21, 2026https://www.theinformation.com/briefings/aws-vp-led-agent-building-service-departs-eight-months