AI News Archive: August 21, 2026 — Part 3
Sourced from 500+ daily AI sources, scored by relevance.
- How AWS Marketplace is using AI agents to meet the rising demand for AI agents
Increasingly, AI agents are handling the nitty-gritty admin and due diligence tasks, but agent-powered marketplaces won't replace live human sales reps or engineers anytime soon.
- Florida seeks court ruling to officially classify Sam Altman and ChatGPT as a 'public nuisance' — OpenAI fights to keep lawsuit away from a state jury
Florida's lawsuit against OpenAI and Sam Altman has now been sitting before U.S. District Judge Aileen Cannon in Fort Pierce for seven weeks.
- China's LLMs Now Lead Global Token Usage for Fifteen Straight Weeks — and DeepSeek-V4-Flash Just Took the Top Spot
OpenRouter data shows Chinese LLMs crossed 34.25 trillion weekly tokens for the first time, with the top four slots all Chinese. DeepSeek-V4-Flash official release vaulted to number one with 570 percent week-on-week growth.
Score: 52🌐 MovesAug 21, 2026https://pandaily.com/china-llm-token-usage-15-weeks-deepseek-v4-flash-leader-aug2026 - The Mistral paradox: Europe’s push for tech sovereignty relies on China’s Z.ai
On August 11, a tiny line buried in a lengthy press release from French company Mistral AI spoke volumes about the strategic and technological dilemmas facing the European Union, as the bloc prepares to navigate an autumn period full of geopolitical and trade upheaval. In a bid to roll out “European infrastructure for sovereign AI”, Mistral – Europe’s great hope in the global artificial intelligence race – said it would begin offering other companies’ models, while keeping the service itself...
- Chinese regulators tell Tesla to fix nearly 3 million cars
Chinese safety regulators have cracked down on doors that don't open in a crash.
Score: 52🌐 MovesAug 21, 2026https://arstechnica.com/cars/2026/08/chinese-regulators-tell-tesla-to-fix-nearly-3-million-cars/ - How AI and power systems are rewriting each other's rules
Data centre electricity demand could double by 2030. AI and power systems are becoming mutually dependent – and both sets of rules are being rewritten.
Score: 52🌐 MovesAug 21, 2026https://www.weforum.org/stories/energy-transition/how-ai-and-power-systems-are-rewriting-each-others-rules/ - Adobe wants Firefly to handle the entire soundtrack for your videos
Adobe Firefly has spent the past few years learning to make images and video. Now Adobe wants it handling the soundtrack too. Generate Music, Generate Speech, and Generate Sound Effects are now generally available in Firefly, giving creators tools for producing music, narration, and custom effects without leaving Adobe’s AI workspace. Adobe says Generate Music […]
Score: 52🌐 MovesAug 21, 2026https://www.digitaltrends.com/computing/adobe-wants-firefly-to-handle-the-entire-soundtrack-for-your-videos/ - Army cyber defenses need ‘dedicated funding’ for AI, top official says
Army cyber defenses need ‘dedicated funding’ for AI, top official says Breaking Defense
Score: 52🌐 MovesAug 21, 2026https://breakingdefense.com/2026/08/army-cyber-defenses-need-dedicated-funding-for-ai-top-official-says/ - ZenMux Opens Free Trial for GLM 5.3
ZenMux Opens Free Trial for GLM 5.3 USA Today
Score: 52🤖 ModelsAug 21, 2026https://www.usatoday.com/press-release/story/40872/zenmux-opens-free-trial-for-glm-5-3/ - Trump says any governor or mayor should want to welcome an AI data center
Trump continued to be bullish on artificial intelligence during a Wednesday speech, during which he urged governors and mayors to accept data centers in their communities.
Score: 52🌐 MovesAug 21, 2026https://www.foxbusiness.com/politics/trump-says-any-governor-mayor-should-want-welcome-ai-data-center - Global investors use 'perp' bets to chase China tech stocks like Unitree
Global investors use 'perp' bets to chase China tech stocks like Unitree Nikkei Asia
Score: 52🌐 MovesAug 21, 2026https://asia.nikkei.com/business/markets/global-investors-use-perp-bets-to-chase-china-tech-stocks-like-unitree - ChatGPT wants Full Disk Access to your Mac and Messages
<>&<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><&&><><><><><>Apple wasn’t joking when it warned us that competitors want access to our most private data. Now, OpenAI’s ChatGPT has introduced a new plugin that can control Apple’s Messages app on Macs. This follows the earlier introduction of a new Computer History feature that monitors what you do on your Mac . And while I don’t think Apple will challenge the new feature — yet — on the Mac, I do see trouble ahead on other platforms. So, what’s ChatGPT’s new thing? Its latest tool means you can use the tool to read, write and send texts via Messages. The app can also search through messages and get message summaries and other information directly from Messages. What is happening? “The Apple Messages plugin is available on all plans in the ChatGPT desktop app for macOS,” says OpenAI . “In Codex and ChatGPT Work, it can read and search iMessage, SMS, and RCS chats on your Mac and send messages on your behalf through the Messages app. It doesn’t let you interact with ChatGPT remotely through Messages, and it doesn’t work in regular ChatGPT chats.” So, if you’ve ever wanted an AI system to go through all your messages to pluck out insights, you’re in luck (though you’ll probably want to check in with corporate legal before doing so). The system does require user consent before working, and OpenAI itself suggests such permission is given only on a per-use, rather than a blanket basis. How might you use this tool? ChatGPT’s tool is not dissimilar to the contextual personal data insights promised by Apple’s SiriAI. You might use it to search for information buried in old messages, create new messages, delete them, and more. You might get ChatGPT to recommend a set of secure browser and messaging services you could use without sharing your information with an AI company and ask it to write a cheery message about dystopia for you, for example. There’s an ad featuring much more mundane examples here . The company has suggested some prompts, including: Suggest follow ups from yesterday in messages. Check my calendar and reply to (contact name) with a few times I’m free for dinner next week. Find birthdays in Messages and add them to my calendar. How does it work? The way the system works isn’t quite as clear as I’d like it to be, given the personal data being parsed, along with recent history on how ChatGPT protects such data on Macs. OpenAI told Bloomberg the plug-in runs locally on the Mac, and doesn’t create an index of a user’s messages. But that may not mean much, given the system does read a user’s existing messages and presumably has some kind of record of the data analysis itself. It is also not especially reassuring that OpenAI tells people not to turn on persistent approval, saying that doing so, “removes your final chance to review a message before ChatGPT sends it as you.” Perhaps of more concern is that the plugin also demands Full Disk Access in System Settings, along with access to contact names and automation tools. Something to think about The degree to which OpenAI is digging into the macOS to make these features work is already driving some backlash. Some apologists tend to dismiss concerns around privacy , while others warn that AI automation is becoming an always-on surveillance system that itself becomes a target for exploitation and attack. Both sides may have a point. But what I don’t see yet is any clear transparency around how the system delivers all its promised convenience without undermining user privacy. It’s all well and good to require consent to make the AI magic happen, but it would be far better, and more legitimate, for such consent to be informed. It also seems very likely OpenAI plans to bring similar features to iPhones and iPads, which might yet open up a new front in Apple/OpenAI’s ongoing litigation around the provision of equal access to user data. That’s particularly true in Europe, where the Digital Markets Act requires Apple to provide third-party AI developers with the same deep system-level APIs and device access that Apple uses for its own AI tools. It remains to be seen how Apple intends to meet that commitment, particularly as it has chosen not to offer SiriAI in Europe until it can agree on some way to offer that kind of access to third parties while continuing to protect user privacy. It is also hard to ignore that ChatGPT on a Mac has now become a prime target for hackers; if they can’t get into Messages directly, now all they need to do is hack the ChatGPT app to do it for them — perhaps using ChatGPT to help them build the code with which to do it . What’s next? I don’t think Apple will challenge OpenAI’s approach for now, as the company doesn’t appear to have actively subverted Mac security protection. But I do predict a showdown on iOS, particularly if Apple is unable to build support for the “ trusted intermediary ” approach it originally proposed to the EU , if only because of the wider extent of information collected on mobile. It also puts yet another slant on the ongoing litigation between both companies . Join me on BlueSky , LinkedIn , Mastodon and subscribe to my newsletter for news and analysis.
Score: 52🌐 MovesAug 21, 2026https://www.computerworld.com/article/4212509/chatgpt-wants-full-disk-access-to-your-mac-and-messages.html - IDC on the Ground: AMD Advancing AI 2026 – Positioning Desktop AI Development as a Foundation for Scalable AI Deployment
At AMD Advancing AI 2026, held on July 23, 2026, at the Moscone Center in San Francisco, California, AMD presented a broad AI strategy that spans cloud infrastructure, enterprise AI, client computing, robotics, and embedded systems. The event brought together developers, customers, ecosystem partners, and enterprise leaders to discuss how AMD’s end-to-end AI portfolio is […] The post IDC on the Ground: AMD Advancing AI 2026 – Positioning Desktop AI Development as a Foundation for Scalable AI Deployment appeared first on IDC .
Score: 51🌐 MovesAug 21, 2026https://www.idc.com/resource-center/blog/idc-on-the-ground-amd-advancing-ai-2026/ - What retail investors need to know before jumping into Anthropic’s IPO
Don’t rush in, there are many business risks for the maker of Claude
Score: 50💰 MoneyAug 21, 2026https://www.ft.com/content/a301daae-05dd-4cf9-b78c-7879e4da55ba?syn-25a6b1a6=1 - Ranking AI Labs on Safety, AT&T’s Open-Source Pivot, Stripe to Buy AI Marketplace OpenRouter — TITV [Video]
Ranking AI Labs on Safety, AT&T’s Open-Source Pivot, Stripe to Buy AI Marketplace OpenRouter — TITV [Video] The Information
- China and US push Southeast Asia over their AI blocs. Will it test region’s non-alignment?
Southeast Asian leaders have long insisted they will not be forced to pick sides in the great-power rivalry between the US and China. But their posture is about to be further tested. This time, the battleground is AI. While Washington wants the region locked into Pax Silica – its bid to build a China-free artificial intelligence supply chain – Beijing wants its neighbours to join China’s own World Artificial Intelligence Cooperation Organisation (Waico). Sources told the South China Morning Post...
- Apple Cuts Jobs in Siri, Vision Pro Immersive Video and Gaming Teams
Apple Inc. is cutting jobs across teams responsible for the Siri digital assistant and the Vision Pro headset, part of an effort to focus on new devices and artificial intelligence.
- Politics hits data centers, OpenAI falls behind Anthropic and now AI is too big to fail… quietly
Data centers, of all things, now look like they’re going to be a prime political issue in the midterm elections and beyond. Really? Really. Even the GOP is worried that opposition to AI data centers could give Democrats a potent campaign issue. It seems a little odd given that data centers are decades old, power […] The post Politics hits data centers, OpenAI falls behind Anthropic and now AI is too big to fail… quietly appeared first on SiliconANGLE .
- Maximizing AI Factory Performance per Watt with NVIDIA DSX MaxLPS
AI factories are power-constrained industrial systems. The question is no longer how many GPUs fit in a data center, but how much AI output each available...
Score: 49🌐 MovesAug 21, 2026https://developer.nvidia.com/blog/maximizing-ai-factory-performance-per-watt-with-nvidia-dsx-maxlps/ - Google Pixel 11 Boot Screen Drops ‘Powered by Android’ for Gemini, and the Symbolism Is Huge
In recent years, Google has increasingly been positioning its flagship Pixel phones around Gemini, its multimodal artificial intelligence chatbot. As the Mountain View-based tech giant continues to push AI across more of its products and services, the effect is now being strongly observed, especially on the new Pixel 11 series. There appears to be a noticeable change ...
- Flock Is Quietly Selling Powerful Drones That Scan License Plates From the Sky
It's the company's "fastest growing business unit." The post Flock Is Quietly Selling Powerful Drones That Scan License Plates From the Sky appeared first on Futurism .
Score: 48🌐 MovesAug 21, 2026https://futurism.com/future-society/flock-quietly-selling-drones-scan-license-plates - Complex odors prove easier to map than expected with machine learning
If you want to describe a particular color, you could look to the Pantone color wheel to find its exact hue, saturation and brightness, and how it compares with other colors. But nothing like that has existed for complex odors.
- China’s telecoms giants bet on ‘token factories’ as AI drives revenue growth
China’s three state-owned telecoms giants have made billable artificial intelligence tokens a key focus of their growth strategies, as AI adoption accelerates across the country and computing demand soars. China Mobile, China Telecom and China Unicom all pointed to AI and computing operations as key growth drivers in their first-half financial disclosures released this month, presenting usage of tokens – the fundamental unit of data processed by AI models – as a central metric alongside others...
- From Atari to EVE Online: Building on 15 Years of AI Research in Games
Google DeepMind partners with game studios to prototype breakthrough AI gameplay.
Score: 48🌐 MovesAug 21, 2026https://deepmind.google/blog/from-atari-to-eve-online-building-on-15-years-of-ai-research-in-games/ - Alibaba Launches CosyVoice Studio, Its First Full-Stack Voice AI Platform, and Bets Speech Is the Next Productivity Entry Point
CosyVoice Studio bundles Alibaba's Qwen-Audio speech family, including the Artificial Analysis top-ranked Qwen-Audio-3.0-Realtime, into one platform with three product modules. Alibaba is positioning speech as the next agent entry point, alongside text and vision.
Score: 48🌐 MovesAug 21, 2026https://pandaily.com/alibaba-cosyvoice-studio-full-stack-voice-ai-platform-aug2026 - Intelligence Indeed’s Z-Agent Becomes OSWorld’s First 90%+ Dark Horse
Intelligence Indeed’s Z-Agent Becomes OSWorld’s First 90%+ Dark Horse USA Today
- Paytm to step up AI, financial services investments after first profitable year, Sharma says
The comments come after a sharp turnaround at One 97 Communications. Paytm reported its first full-year profit of Rs 552 crore in FY26, compared with a loss of Rs 663 crore a year earlier, while operating revenue rose 22% to Rs 8,437 crore. Ebitda improved by Rs 2,008 crore and hit Rs 502 crore.
- Exclusive: Key OpenAI Sales Executive Kaylin Voss Returns to Salesforce
Exclusive: Key OpenAI Sales Executive Kaylin Voss Returns to Salesforce The Information
Score: 48🌐 MovesAug 21, 2026https://www.theinformation.com/briefings/exclusive-key-openai-sales-executive-kaylin-voss-resigns - AI Data Centers Would Go Dark First On North America’s Largest Grid
PJM would cut new 50-megawatt data centers before anyone else unless they bring newly built power. FERC told every US grid to do the same.
- Nvidia finds that simple linear math can replace costly AI model handoffs
When an agentic AI system hands a task from a small model to a larger one — or back down again — it pays a steep tax: the receiving model has to recompute the entire conversation from scratch, driving up compute costs and latency. This is a major bottleneck for enterprises building long-horizon, multi-LLM workflows. To solve this challenge, researchers at Nvidia have introduced a cross-model KV cache transfer technique that directly maps the prefilled KV cache from a source model into the target model. This technique aligns with real-world agentic applications where large contexts accumulate across many turns. For real-world AI applications, cross-model KV cache transfer can reduce compute costs and latency on long-running, multi-LLM workflows — and it does so with simple linear math, not an expensive deep learning model. Experiments show that, on compatible model pairs, this linear mapping process runs 2.7 to 25 times faster than recomputing the conversation while retaining up to 98% of the target model's standalone accuracy. Why swapping models mid-session is so expensive Examining how LLMs handle memory helps understand why multi-model workflows hit a performance wall in production. When an LLM receives a prompt, it must first execute the “prefill” stage, which is the initial forward pass that computes the keys and values for all input tokens and populates the Key-Value (KV) cache. After that, it enters the “decode” phase, where it computes and generates the next tokens in the sequence. During this phase, the model reads from this KV cache to predict new tokens one by one, bypassing the need to re-evaluate the entire history of the conversation for each new token. In multi-turn conversations or long-horizon agentic sessions, the context gradually becomes longer. Because the computational cost of the prefill stage scales directly with both model size and input length, processing these long sessions becomes increasingly expensive and introduces significant latency if the KV cache is invalidated. This invalidation happens whenever the AI system tries to swap models mid-session, such as routing a complex reasoning step to a larger model or dropping to a smaller model to save costs. Because different LLMs have different architectures, they expect their cache inputs in different formats. As a result, any model switch forces the receiving model to repay the entire prefill cost from scratch to recompute the KV cache for the accumulated context. Mapping memory between models without starting over The Nvidia researchers studied cross-model KV cache transfer to see how developers can transform the KV cache of one model into the expected format of another without running the prefill phase again. If solved, cross-model KV cache transfer has benefits in both directions. Small-to-large model transfer upgrades the quality of the output. For example, a cheap, small model handles the routine parts of an agentic workflow but struggles with a complex reasoning problem, and you map the KV cache to a larger model and continue the process seamlessly. On the other hand, large-to-small model transfer reduces compute costs. A highly capable, large model might be used to unpack a massive, complex system prompt or synthesize a dense PDF at the start of a session. Once the heavy lifting is done, the session's KV cache is mapped down to a smaller, more economical model to handle the rapid-fire, conversational turns that follow. There have been previous efforts to solve the KV cache transfer problem, but they suffer from a few key limitations. These include the need for expensive gradient-based training or very strict architectural constraints. For this initial study, the authors restricted their focus to within-family transfers, such as transitioning between different-sized models in the Qwen, Llama, or Ministral families. These models share tokenizers, training data DNA, and core architectural styles but differ in size and depth. However, this framework leaves plenty of room for future experiments. The researchers note the technique could eventually be expanded to cross-family transfers, mismatched KV head counts, or hybrid architectures that blend standard attention with other memory mechanisms. The key finding of the Nvidia study is that cross-model KV cache is a significantly linear structure. This means you can do the mapping with simple algebra tricks and without the need for heavy neural network training. For example, when experimenting on KV cache transfer from a 14-billion parameter Qwen3 model to a 32-billion parameter version, the authors discovered that a simple linear regression mapping from one source layer to a target layer can recover 56% of the variance in the target’s keys and 32% of the variance in its values. When combining multiple source layers, those numbers climbed to 79% and 65% respectively. To translate this linear relationship into a practical system, the researchers designed a closed-form per-head ridge mapper with three key components: Per-head ridge regression: Instead of using complex deep learning to train the system, they fit a simple linear regression using a tiny calibration set of a few hundred text sequences. This technique solves a classic line-of-best-fit problem independently for every attention head. Cross-layer source selection: Because the source and target models have different numbers of layers, the mapper evaluates and selects the most predictive source layers to feed into each specific target layer. This way, the system picks only the most helpful pieces of memory from the old model to construct the new model's memory. Content-space mapping: Before translating the data, the mapper strips away the RoPE encodings. RoPE, or Rotary Position Embedding, is a standard mechanism that applies a mathematical, position-dependent rotation to the data so the model understands the order of the tokens in a sequence. Stripping the RoPE values makes it possible for the mapper to generalize to sequences of lengths larger than its training data. Putting the linear mapper to the test To test whether the technique works, the researchers evaluated the transfer pipeline across six “matched-KV” model families. Matched-KV means the source and target models share the same KV head count and per-head dimensions, which is typical for different-sized models within the same family. The model families included Qwen3, Llama 3.1, and Ministral 3, with tests for KV cache transfer across different sizes ranging from 3 billion to 70 billion parameters. Their experiments included a massive 8.8x parameter leap from Llama 3.1 8B to 70B. To cover a wide range of tasks, they evaluated the models on five core accuracy benchmarks (ARC-Challenge, HellaSwag, WinoGrande, MMLU, and GSM8K) as well as language modeling perplexity on WikiText-2 and a multi-turn conversation task called CoQA. To fit the linear translation mapper, they used a tiny calibration dataset of just 500 text sequences of 1,024 tokens each. The researchers compared the framework against the baseline ceiling accuracy where the target model does a full, traditional prefill. They also compared their full system against ablated configurations, such as reducing the number of selected layers or deactivating different components. Additionally, they compared their simple method against a deep neural network trained with backpropagation to see if heavier deep learning could recover accuracy on pairs where the linear method struggled. For four of the six tested pairs, the fast, closed-form linear ridge mapper retained 73% to 98% of the target's standalone prefill accuracy — including the massive leap from Llama 3.1 8B to 70B, which retained 72.8% of target accuracy. The mapper also runs between 2.7 and 25 times faster than re-prefilling. For example, when translating a 32,768-token KV cache from a Qwen3 14B to a 32B model, the transfer took just 278 milliseconds, compared to nearly 7 seconds for a standard re-prefill. The system also demonstrated high stability on tasks that run across many steps. When tested on multi-turn conversations, the drift, or accuracy loss, between the target baseline and the transferred cache remained incredibly small across 10 turns, proving it will not cascade into failure during long agentic sessions. However, the straightforward linear approach did run into limitations on specific model pairs. For two of the Ministral configurations, the linear mapper degraded sharply because the simple linear fit failed to extrapolate outside calibration data. To fix this, the researchers swapped the linear mapper for a nonlinear multi-layer perceptron (MLP) with two 1,024-unit hidden layers trained on the same data. This added a complexity and training tax to the setup, but it recovered their accuracy to above 90%. A bigger industry problem than one paper can solve The introduction of cross-model transfer is part of a broader, industry-wide push to solve the KV cache bottleneck, which has emerged as one of the key hurdles for scaling enterprise AI. As developers push LLMs to process massive documents or code bases and execute long-running reasoning tasks, managing this memory layer is becoming as important as the models themselves. Over the past year, researchers have attacked this compute and memory problem from multiple angles. For instance, Nvidia recently introduced dynamic memory sparsification (DMS), a technique that intelligently evicts less important tokens from the KV cache to cut reasoning costs by up to 8x. Other approaches focus on aggressive data compression. MIT researchers developed an algebraic compaction technique called Attention Matching that compresses the KV cache by 50x without degrading quality. Similarly, Nvidia introduced KV Cache Transform Coding (KVTC), which borrows media compression concepts to shrink memory by 20x without altering the underlying model weights. Beyond compression, researchers are also attacking the computational overhead of memory retrieval. Optimizers like IndexCache strip away redundant layer calculations to deliver significantly faster time-to-first-token in long-context applications. And models like DeepSeek and the GLM series are optimizing the KV cache through architecture innovations. As AI systems take on longer-horizon tasks and more complex architectures, the underlying memory infrastructure is becoming as important as the models themselves. Cross-model KV cache transfer gives developers one more tool for keeping inference costs down as they scale multi-model agentic systems.
Score: 48🌐 MovesAug 21, 2026https://venturebeat.com/technology/nvidia-finds-that-simple-linear-math-can-replace-costly-ai-model-handoffs - As demand for Meta AI glasses explodes, it’s harder to avoid creepy recordings
Ars looks at Zuckoff, the latest free app detecting Meta AI glasses amid privacy backlash.
- MAHA warns Trump against coal-powered AI data centers
Prominent figures in the Make America Healthy Again movement are telling President Trump that "data centers should not become the justification for burning more coal." Why it matters: The appeal, in a joint letter , collides with Trump officials' support for coal to help power AI. It cites coal's public health and environmental toll due to soot pollution, mercury, and byproducts called "coal ash" that contain heavy metals. The big picture: The letter signals an escalation of MAHA engagement on data centers and what's powering them. It urges the administration to convene MAHA leaders, farmers, physicians and others to craft "health-first standards for AI infrastructure." It says there are better energy options, citing solar paired with energy storage, advanced geothermal, and improved energy efficiency. Driving the news: The many signers of the letter include ... Moms Across America founder Zen Honeycutt. Charles Eisenstein, who was a speechwriter and adviser on Health Secretary Robert F. Kennedy Jr.'s 2024 presidential campaign. Aimee Villella McBride, executive director of the Global Wellness Forum. David Murphy, founder and CEO of United We Eat. In addition to President Trump, the joint letter is copied to Kennedy, Environmental Protection Agency head Lee Zeldin, Energy Secretary Chris Wright, and Agriculture Secretary Brooke Rollins. State of play: The Trump administration is moving on multiple fronts to boost coal-fired power, often citing AI power demand among the reasons. An April 2025 executive order calls coal "critical" to meeting rising power demand from data centers and other needs. The Energy Department has announced hundreds of millions of dollars in funding to retain and even expand coal-fired power. What they're saying: The rise of data centers and what's fueling them is a new area of interest for Moms Across America and many in the MAHA movement, Honeycutt said, though she cautions she doesn't speak for the whole movement. "MAHA is not just about ... getting pesticides out of our food, or having vaccines be safe. We're about not having toxins impact the health of our families and our communities and farmers," she said in an interview. "This is a an extremely important issue that is emerging in a new way for the MAHA movement," she said. Catch up quick: The letter arrives as data centers face a growing backlash, and politicians from both parties are floating new restrictions on the industry. The bottom line: The U.S. can lead on AI, energy, and security "without asking children to carry more mercury in their bodies, more soot in their lungs, more heavy metals in their soil and food, or more contamination in their water," the letter states.
Score: 47🌐 MovesAug 21, 2026https://www.axios.com/2026/08/21/ai-data-centers-coal-trump-rfk-maga-warning - SOP-Bench: A new benchmark for evaluating AI agents on real business procedures
Extendable framework enables testing agents on the full set of capabilities required to successfully complete a procedure, not isolated proxy tasks.
Score: 47🌐 MovesAug 21, 2026https://www.amazon.science/blog/sop-bench-a-new-benchmark-for-evaluating-ai-agents-on-real-business-procedures - Nvidia and Korea’s Rebellions reportedly hold partnership talks
Nvidia has poured more than $63bn in investments into other companies. Read more: Nvidia and Korea’s Rebellions reportedly hold partnership talks
Score: 47🌐 MovesAug 21, 2026https://www.siliconrepublic.com/business/nvidia-and-koreas-rebellions-reportedly-hold-partnership-talks - How Silicon Valley just might save America from the growing drone threat
In June, Allen Control Systems raised $200 million on the strength of its Bullfrog platform, which can respond to a drone threat in seconds by shooting them out of the sky with bullets. Read More
Score: 46🌐 MovesAug 21, 2026https://fortune.com/2026/08/21/silicon-valley-just-might-save-america-growing-drone-threat/ - A roadmap for safeguarding against AI bioweapons
AI-enabled bioweapons are a potentially catastrophic yet manageable risk — if government, the scientific community, the public health sector and leading tech companies can develop appropriate safeguards, a new report argues. Why it matters: The debate over AI and public safety isn't one that the health care or research communities can ignore. Driving the news: A RAND report out this week outlines nine mitigation strategies targeting a range of actors who could use AI to design and release a biological weapon. While there's already considerable debate about government oversight and safeguards around frontier AI companies, RAND calls for more controls and a wide range of cross-industry cooperation. It recommends protecting the same biological tools and datasets that are contributing to some of today's most exciting medical advances. And though we all think of public health preparedness as a response to an outbreak, the report frames it as a form of deterrence. The big picture: The nine measures would prevent a large-scale AI-enabled biological attack by managing access to information that can fall into the wrong hands, disrupting access to the materials needed to make a weapon, proactively detecting signs of misuse and deterring nefarious activity from occurring in the first place. Part of the problem with AI's rapid advancement is it expands the field of players potentially capable of making a biological weapon, from lone wolves to state-backed actors. "AI capabilities in biology are advancing really fast, but luckily most dangerous thresholds have not yet been crossed," RAND lead author Steph Guerra told me. Given the nature of the threat, "we need this layered system of mutually reinforcing approaches." Zoom in: Remember all of that talk about AI helping discover drug candidates? Some of the technology enabling that work could turn dangerous in the wrong hands, Guerra told me. A lot of the tools and biological data being generated by pharmaceutical companies and biotechs are proprietary, meaning access is already restricted. Open-source biological tools and data are more concerning — including those emerging from academia and research institutes. "Right now, data for biology is freely out there and open because biology ... [is] for discovery and saving lives. It's not warfare first, which is a good thing," Guerra said. Yes, but: Restricting what's publicly accessible risks slowing down scientific progress. "Let's build big moonshot projects and biodata factories that produce the data we need to cure diseases, but we can also create tiered systems of access so legitimate scientists are the ones who access the most misuse-relevant datasets," Guerra told me. The intrigue: The RAND team has been exploring how the use of AI agents can make biological developments more widely accessible, lowering the bar for the level of expertise needed to make bioweapons. Just this week, Anthropic announced that Claude successfully designed protein binders against 14 targets — an important step in the drug development process. In a blog post, Anthropic acknowledged both the promise and the danger, noting that protein design and other research biology capabilities aren't available for general access in the company's most capable model. "The uplift provided by the increasingly autonomous research capabilities of AI models will undoubtedly speed the development of human therapies and fundamental scientific discoveries," the company wrote. But, it added, "without robust safety measures, they could enable bad actors to perform dangerous research, such as the development of bioweapons."
- Plymouth-based Oshen raises €4.27 million to scale robot swarms that act as the ocean’s “eyes and ears”
Oshen, a Plymouth-based robotics company building a persistent sensing layer for the world’s oceans, has secured €4.27 million ($5 million in a fresh funding round and is tripling its manufacturing rate to meet demand from navies, weather agencies and ocean scientists. The round was led by Lunar Ventures, with participation from AlbionVC, Twin Track and […] The post Plymouth-based Oshen raises €4.27 million to scale robot swarms that act as the ocean’s “eyes and ears” appeared first on EU-Startups .
- Skilled labor demand is exploding. AI may be both a cause and a solution
A survey by the National Fire Protection Association found that while AI is putting pressure on an already stretched fire and safety workforce, it is also making their jobs easier.
Score: 45🌐 MovesAug 21, 2026https://www.hrdive.com/news/skilled-labor-demand-is-exploding-ai-is-both-a-cause-and-a-solution-nfpa/828486/ - Vantage and Nebius move first in South Wales AI Growth Zone deployment
Vantage and Nebius move first in South Wales AI Growth Zone deployment IT Pro
Score: 45🌐 MovesAug 21, 2026https://www.itpro.com/infrastructure/vantage-and-nebius-move-first-in-south-wales-ai-growth-zone-deployment - Nvidia’s Ruthless Way to Beat the Competition Has 2 Problems
Nvidia’s Ruthless Way to Beat the Competition Has 2 Problems Barron's
- The AI Industry Has Fumbled Its Public Image So Badly That It’s Incredible
The majority of young adults in the US say they're "more concerned than excited" about AI. The post The AI Industry Has Fumbled Its Public Image So Badly That It’s Incredible appeared first on Futurism .
- AI startup linked to OpenAI, Anthropic hacks releases post-incident report: Key findings
AI startup linked to OpenAI, Anthropic hacks releases post-incident report: Key findings
- Anthropic Claude Adds Watermarks. Implications For Business?
Anthropic just watermarked all Claude text for EU rules. What this precedent means for your business, and specific actions that business leaders can take today.
- China has more than 70 operational embodied-AI training grounds, report says
More than 70 embodied-AI training grounds had been built and put into operation across China by the end of June, according to a 2026 report from the China Academy of Information and Communications Technology. Another 46 facilities were under construction or in the planning stage, with the sites spread across more than half of China’s […]
Score: 45🌐 MovesAug 21, 2026https://technode.com/2026/08/21/china-has-more-than-70-operational-embodied-ai-training-grounds-report-says/ - AI-generated ads outperform human designers in live campaign, advantage holds 18 months later
The common belief throughout the marketing and advertising world is that generative AI can produce striking visuals, but it struggles to match the performance and brand discipline of skilled human designers. New research published in the journal Marketing Science challenges that assumption, finding that a carefully designed AI system can outperform visuals created by humans, and it can outperform designs created by another AI model that was optimized solely on the basis of aesthetics.
Score: 45🌐 MovesAug 21, 2026https://techxplore.com/news/2026-08-ai-generated-ads-outperform-human.html - Schools are starting to teach AI literacy. For many, that means helping kids see chatbots' flaws
AI chatbots have become the bane of teachers everywhere
Score: 45🌐 MovesAug 21, 2026https://abcnews.com/Technology/wireStory/schools-starting-teach-ai-literacy-means-helping-kids-135828390 - Vietnam to include artificial intelligence in school curriculum
Vietnamese school students will receive 12 periods of artificial intelligence (AI) education each academic year under a framework recently issued by the Ministry of Education and Training.
- Private Equity Is Deploying an Army of AI Wonks to Embed in the Firms They Back
Wall Street firms see big business potential in offering AI services to the broader business world.
- Stockholm-based Eyedentity secures €1.3 million to detect deadly eye cancer earlier with AI
MedTech startup Eyedentity has closed its first oversubscribed investment round at €1.3 million (SEK 15 million) to support regulatory approval, further product development and the expansion of Eyedentity’s AI-powered diagnostics platform into opticians, eye clinics and hospitals. The round was led by Norrsken Launcher, with participation from the Karolinska Institutet Innovation and a group of […] The post Stockholm-based Eyedentity secures €1.3 million to detect deadly eye cancer earlier with AI appeared first on EU-Startups .
- Chinese humanoid robots' biggest obstacle: Humans are still (mostly) better
Humanoid robots still struggle to perform as efficiently as humans in most labor scenarios.
Score: 44🌐 MovesAug 21, 2026https://www.cnbc.com/2026/08/21/chinese-humanoid-robots-face-challenge-of-their-own-capabilities.html