AI News Archive: August 21, 2026 — Part 1
Sourced from 500+ daily AI sources, scored by relevance.
- Nevada allows Uber, Tesla and Waymo to start paid robotaxi service
Nevada's transportation authorities approved the companies' application to charge for robotaxi rides.
Score: 91🌐 MovesAug 21, 2026https://www.engadget.com/2241379/nevada-allows-uber-tesla-waymo-paid-robotaxis/ - Anthropic Aims to Match or Top SpaceX IPO
Anthropic is aiming for an IPO that could make history, expecting to match or even top SpaceX's record-setting debut, according to sources. The maker of Claude could file publicly by the end of this month. The company raised $65 billion in May at a $965 billion valuation. Bloomberg's Shirin Ghaffary, who leads coverage of Anthropic, and Bloomberg's Liana Baker, who leads the deals team, join Ed Ludlow on "Bloomberg Tech." (Source: Bloomberg)
Score: 90💰 MoneyAug 21, 2026https://www.bloomberg.com/news/videos/2026-08-21/anthropic-aims-to-match-or-top-spacex-ipo-video - How AI accounting startup Rillet raised $100M and became a unicorn in 48 hours
Rillet CEO Nicolas Kopp shared growth numbers at a board meeting and set off a fundraising frenzy from Iconiq, Sequoia and others. Without even trying.
Score: 89💰 MoneyAug 21, 2026https://techcrunch.com/2026/08/21/how-ai-accounting-startup-rillet-raised-100m-and-became-a-unicorn-in-48-hours/ - Uber fined nearly $1B by Dutch regulators over automated suspensions of accounts
Dutch data protection authorities have fined Uber nearly $1 billion, saying the ride-hailing company used automated software to suspend driver accounts, sometimes permanently with no human review to check for mistakes
Score: 88🌐 MovesAug 21, 2026https://abcnews.com/Technology/wireStory/uber-fined-1-billion-dutch-regulators-automated-suspensions-135848560 - US wants to force partner countries to choose between Washington and Beijing in the AI race
The US is drafting a letter to partner countries telling them to pick a side in the AI standoff between Washington and Beijing, according to Reuters. The article US wants to force partner countries to choose between Washington and Beijing in the AI race appeared first on The Decoder .
- ChatGPT on Mac can now read and respond to Apple iMessages
OpenAI has rolled out a plugin that allows ChatGPT to control the Apple Messages app.
- China's First 100,000-Card Domestic AI Super-Cluster Goes Live, and the National Compute Grid Takes Shape
A fully domestic 100,000-card AI super-cluster came online this month at the National Supercomputing Internet Zhengzhou core node, with a parallel deployment in the Greater Bay Area. The NDRC is now framing the next five-year plan around a single national compute fabric.
Score: 85🌐 MovesAug 21, 2026https://pandaily.com/china-first-100k-card-domestic-ai-supercluster-national-compute-grid-aug2026 - Micron bets in $10B in AI memory lab
Micron invests $10B in AI memory lab, boosting AI hardware capabilities.
- Qwen3.8-27B Is the New Model Kill Line: Local 'Opus 4.6' on Consumer Hardware
Open-source developers have a new benchmark of their own: Qwen3.8-27B. Released August 14, it passed a million downloads in two days, hit Hugging Face's global trend list, and earned 52 on Artificial Analysis' Intelligence Index, an interval shared with GPT-5.6 Luna and DeepSeek V4 Flash. Nicknamed 'local Opus 4.6', it squeezes frontier-level capability into a 27-billion-parameter model that fits a 24GB GPU after 4-bit quantization. Now every new release is asked: is it better than Qwen3.8-27B?
Score: 83🤖 ModelsAug 21, 2026https://pandaily.com/qwen3-8-27b-new-model-kill-line-local-opus-consumer-hardware-aug2026 - $1bn AWS engineering firm to accelerate African AI
The new business will help African companies overcome the gap between AI experimentation and deployment, says AWS.
Score: 82💰 MoneyAug 21, 2026https://www.itweb.co.za/article/1bn-aws-engineering-firm-to-accelerate-african-ai/JBwEr7n3WOEM6Db2 - Anthropic Plans to One-Up SpaceX in Race for AI Dominance
Anthropic Plans to One-Up SpaceX in Race for AI Dominance Barron's
- Anthropic Preps for Blockbuster Public Listing | Bloomberg Tech 8/21/2026
Bloomberg’s Ed Ludlow takes a look at Anthropic as the AI company expects to match or top the size of SpaceX's record-setting IPO. Plus, sources say Broadcom is in talks to raise more than $60 billion in debt to help Anthropic and others secure chips and computing power, and all eyes are on Nvidia as the chipmaker reports earnings next week. (Source: Bloomberg)
Score: 81💰 MoneyAug 21, 2026https://www.bloomberg.com/news/videos/2026-08-21/bloomberg-tech-8-21-2026-video - Stripe agrees to acquire OpenRouter for $8 billion
With a deal, OpenRouter’s business could be destined for commoditization, and that’s actually kind of the point.
Score: 81💰 MoneyAug 21, 2026https://www.semafor.com/article/08/21/2026/stripe-agrees-to-acquires-openrouter - OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20%
OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20% Reuters
- Broadcom seeks $60 billion to fund the next AI infrastructure boom
Broadcom seeks $60 billion to fund the next AI infrastructure boom YourStory.com
- Waymo builds its own chip for its robotaxis, cutting its reliance on Nvidia
Waymo built its own chip for its robotaxis, cutting its reliance on Nvidia. The article Waymo builds its own chip for its robotaxis, cutting its reliance on Nvidia appeared first on The Decoder .
Score: 81🌐 MovesAug 21, 2026https://the-decoder.com/waymo-builds-its-own-chip-for-its-robotaxis-cutting-its-reliance-on-nvidia/ - Nvidia is acquiring Poolside's "Model Factory" and 109 employees for $6 billion
Nvidia is paying $6 billion for software that builds AI models from the startup Poolside, and it wants to bring on 109 employees. The article Nvidia is acquiring Poolside's "Model Factory" and 109 employees for $6 billion appeared first on The Decoder .
Score: 81💰 MoneyAug 21, 2026https://the-decoder.com/nvidia-is-acquiring-poolsides-model-factory-and-109-employees-for-6-billion/ - Starcloud raises $250M to support the creation of data center satellite network in league with Nvidia
Seattle-area space startup will use Nvidia's next-gen chip on its own next-gen satellites for AI data processing. Read More
Score: 80💰 MoneyAug 21, 2026https://www.geekwire.com/2026/starcloud-250m-data-center-satellite-network-nvidia/ - H200 AI GPUs finally reach China under case-by-case import licenses, but it's already too late for Nvidia — homemade chips corner the China market as country seeks semiconductor independence
Most of each company's U.S.-licensed allowance, understood to be up to 100,000 units apiece, must stay outside the mainland.
- Anthropic puts its most powerful model Claude Mythos 5 to work for cyber defense
Anthropic is now running its security scanner Claude Security on Claude Mythos 5. The tool scans codebases for vulnerabilities, provides severity ratings with CWE classifications, and suggests patches. Anthropic is also plugging Mythos 5 into partner security products protecting critical infrastructure. The article Anthropic puts its most powerful model Claude Mythos 5 to work for cyber defense appeared first on The Decoder .
Score: 78🤖 ModelsAug 21, 2026https://the-decoder.com/anthropic-puts-its-most-powerful-model-claude-mythos-5-to-work-for-cyber-defense/ - NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents
A frontier language model is only one component of an AI agent. The surrounding agent system—often called a harness—determines how the model receives...
- Google says Gemma has passed a billion downloads
Google says its Gemma family of open models has passed a billion downloads. Developers have published more than 100,000 Gemma variants over the past two years, the company said in a blog post on Thursday. Clement Farabet, a vice president at Google DeepMind, and Olivier Lacombe, a product director there, wrote it. Google calls the […] This story continues at The Next Web
Score: 77🌐 MovesAug 21, 2026https://thenextweb.com/news/google-gemma-one-billion-downloads-gemmaverse-variants - Supermicro fires several employees following investigation into $2.5 billion China AI chip smuggling — claims that senior management had no knowledge of illicit transactions
An independent investigation on Supermicro clears senior management from any wrong-doing and also says that its financial statements were still reliable, despite the alleged diversion of its restricted products. Nevertheless, some of its employees from sales, technical support, and business development departments were fired for failing to 'follow Company policies or the Company’s code of conduct.'
- Gravis Robotics bags $200M, a safer marketplace for teen odd jobs, and Monzo chairman Gary Hoffman to stand down
This week, we tracked more than 45 tech funding deals worth over €684 million and over 5 exits, M&A transactions, rumours, and related news stories across Europe.If email is more your thing, you c...
- Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks
Deepseek has released V4-Flash-Vision-Exp, an experimental multimodal model that adds image understanding to V4-Flash's text capabilities. On the company's own multimodal agent benchmarks, it approaches Opus 4.8 and sometimes beats it. The article Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks appeared first on The Decoder .
Score: 76🤖 ModelsAug 21, 2026https://the-decoder.com/deepseek-releases-experimental-flash-vision-model-that-rivals-opus-4-8-on-agent-benchmarks/ - OpenAI adds an AI safety layer to detect misuse without retaining enterprise data
OpenAI is adding a new safety capability that allows enterprises to detect misuse of its AI systems across multiple interactions without retaining prompts or responses, enabling risk monitoring while preserving its Zero Data Retention (ZDR) commitments. “OpenAI does not retain…prompts or model responses after a request is processed,” the company said in a blog post, describing its ZDR approach. The new system, called Private Safety Processing, is “designed to identify patterns across related interactions without giving OpenAI personnel access to the underlying content.” The capability is being tested with eligible enterprise and API customers and is intended to address a limitation in existing safety controls that evaluate interactions individually, making it harder to detect risks that unfold over time. What does Private Safety Processing do Private Safety Processing is designed to extend existing safety systems by correlating activity across related interactions rather than analyzing each prompt in isolation, according to OpenAI. Under the model, automated systems analyze interactions and generate “a narrowly defined signal indicating the type of activity involved,” instead of exposing the underlying prompts or responses, according to OpenAI. The system can operate whether customer data remains within enterprise-controlled infrastructure or is stored by OpenAI with encryption keys controlled by the customer, the company said. “In both cases, automated systems can identify potential misuse and return limited safety signals without exposing the underlying prompts or responses to OpenAI personnel,” the post added. Why existing safety controls fall short OpenAI said the new capability is intended to address a gap in the detection of AI risks. “The most serious AI safety risks are not always visible in a single interaction,” the company said, noting that harmful intent may only become clear when multiple interactions are viewed together. Such risks include repeated attempts to probe safeguards, coordinated activity across accounts, and misuse that emerges over a sequence of interactions, according to the company. As AI systems take on longer and more complex tasks, evaluating individual prompts in isolation can limit the ability to identify such patterns, OpenAI said in the post. Diverging approaches to AI safety The introduction of Private Safety Processing highlights differing approaches to safety among AI providers. OpenAI said the system is designed to detect misuse patterns across interactions while preserving zero data retention. By contrast, some providers retain customer interaction data for a period of time to support safety monitoring, reflecting a different approach to identifying risks that span multiple requests. Sanchit Vir Gogia, chief analyst at Greyhound Research, said the difference lies in how evidence is handled rather than whether signals are used. “This is a disagreement about how much raw content you need besides a signal you are keeping regardless, rather than privacy against surveillance,” Gogia said. He added that both approaches rely on derived indicators but differ in where investigation data resides. “Anthropic wants enough content to investigate the case. OpenAI wants the customer to hold the case while the provider holds the alarm,” he said. Signal-based detection and verification The system’s reliance on signals rather than direct data access shifts how enterprises verify and investigate incidents, analysts noted. “The architecture is entirely viable. Security has worked from derived indicators for a generation. The difficulty is verification, not feasibility,” Gogia said. He added that detecting behavior across time requires retaining some form of representation. “A system cannot detect behaviour across time unless it remembers something across time,” he said. Private Safety Processing “is privacy-preserving abuse detection. It is not an enterprise forensic record, and OpenAI does not claim it is,” he said. Implications for regulated sectors According to Apeksha Kaushik, senior principal analyst at Gartner, the approach could influence AI adoption in industries with strict data requirements. “Privacy-preserving safety models, such as those employing Zero Data Retention (ZDR), represent an emerging approach that may lower barriers to AI adoption in regulated sectors like financial services and healthcare,” she said. Such models “may help organizations address certain privacy requirements and may align with frameworks such as GDPR and HIPAA, contingent on specific implementation details and regulatory guidance,” she noted. Kaushik added that organizations should evaluate such approaches against their compliance requirements. “Organizations are encouraged to consult with their compliance and legal teams to determine whether such approaches meet their specific regulatory and operational requirements,” she said. OpenAI said enterprises retain control over their data under this model and can investigate alerts using their own systems. Customers can also choose to share relevant data with the company to support investigations or appeals. Analysts say this shifts responsibility toward enterprises. “Zero Data Retention does not remove the forensic burden. It relocates it,” Gogia said. The article originally appeared on CSO .
- YMTC parent seeks $4.9bn Shanghai IPO on AI memory boom
YMTC parent seeks $4.9bn Shanghai IPO on AI memory boom Nikkei Asia
Score: 75💰 MoneyAug 21, 2026https://asia.nikkei.com/business/markets/ipo/ymtc-parent-seeks-4.9bn-shanghai-ipo-on-ai-memory-boom - Anthropic Taps Google Chip Veteran as Part of Push Into Hardware
Anthropic PBC has hired Amir Salek, a founder of the custom chip program at Alphabet Inc.’s Google, as the AI lab lays the groundwork for a push into making its own semiconductors.
- OpenAI's GPT-Image-2 can now generate images without a background
OpenAI is previewing transparent background support for GPT-Image-2 through its API. The alpha channel gets baked in during image generation, which according to OpenAI beats conventional background removal. A single parameter activates the feature. The article OpenAI's GPT-Image-2 can now generate images without a background appeared first on The Decoder .
Score: 74🤖 ModelsAug 21, 2026https://the-decoder.com/openais-gpt-image-2-can-now-generate-images-without-a-background/ - US corporate AI debt surge tests investor limits as fatigue emerges
US corporate AI debt surge tests investor limits as fatigue emerges Reuters
- Anthropic’s enterprise venture has bought its second consultancy in four months
Ode, the enterprise AI company Anthropic set up with Blackstone, has bought an AI consultancy called Casper Studios. Casper helps companies put Claude to work inside software their staff already use, Ode said on Thursday. The announcement gives no price. It is the second consultancy Ode has absorbed in four months. The company Anthropic built […] This story continues at The Next Web
Score: 73🌐 MovesAug 21, 2026https://thenextweb.com/news/anthropic-ode-acquires-casper-studios-enterprise-ai - Anthropic’s Opus 4.6 is a smut-machine
Anthropic forbids its Claude models from generating sexually explicit content. But a series of tests conducted by TechCrunch found that it didn't take much to get past the restriction.
Score: 72🤖 ModelsAug 21, 2026https://techcrunch.com/2026/08/21/anthropics-opus-4-6-is-a-smut-machine/ - Nvidia is Using Land and Electricity Deals to Lock In Its Hardware Bundle
Nvidia is Using Land and Electricity Deals to Lock In Its Hardware Bundle The Information
- Meta pays Microsoft hundreds of millions a year to rent AI models
Meta is spending hundreds of millions of dollars a year buying access to AI models through Microsoft’s cloud. The company runs trillions of tokens a week through Azure, Bloomberg reported, citing a person familiar with the matter who asked not to be identified. That makes Meta one of Microsoft’s largest AI customers. Both companies declined […] This story continues at The Next Web
Score: 72🌐 MovesAug 21, 2026https://thenextweb.com/news/meta-microsoft-azure-foundry-ai-model-spending - New Anthropic safety system will let customers keep control of data
A 30-day retention policy is still likely to stick around for "Mythos-class" models.
Score: 72🌐 MovesAug 21, 2026https://www.thestack.technology/new-anthropic-safety-system-will-let-customers-keep-control-of-data/ - Anthropic Brings Claude Mythos 5 to Claude Security: Enterprise Teams Get Frontier Vulnerability Scanning Without Direct Model Access
Anthropic Brings Claude Mythos 5 to Claude Security: Enterprise Teams Get Frontier Vulnerability Scanning Without Direct Model Access MarkTechPost
Score: 71🌐 MovesAug 21, 2026https://www.marktechpost.com/2026/08/21/anthropic-brings-claude-mythos-5-to-claude-security/amp/ - Anthropic changes data retention policy after enterprise pushback
Anthropic is easing its controversial data storage policy, letting enterprise customers keep their own data going forward. The article Anthropic changes data retention policy after enterprise pushback appeared first on The Decoder .
Score: 69🌐 MovesAug 21, 2026https://the-decoder.com/anthropic-changes-data-retention-policy-after-enterprise-pushback/ - Alibaba Releases Qwen-UI-Agent: Making Models Truly Use Every Screen
On August 20, Alibaba Qwen released Qwen-UI-Agent, a GUI agent foundation model for real devices covering phones, desktops, web, and deep search. Trained and evaluated on real hardware, it reads screens and mimics clicks, input, and swipes, scoring 82.1% on MobileWorld, 79.5% on OSWorld-Verified, and top in browser WebArena, while stopping to ask for confirmation before sensitive actions like payments.
Score: 68🤖 ModelsAug 21, 2026https://pandaily.com/alibaba-qwen-ui-agent-gui-agent-foundation-model-screen-control-aug2026 - Nvidia scales back financing guarantee for OpenAI data center
Nvidia is scaling back a proposed financial guarantee tied to a massive OpenAI data center project in Ohio, reducing its initial commitment from as much as $250 billion to less than $120 billion, according to report in the Wall Street Journal . Earlier this month, Nvidia announced partnerships with major financial firms including Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs and KKR, aimed at mobilizing more than $500 billion in capital for AI computing infrastructure. The change represents a significant restructuring of Nvidia’s role in financing the planned facility, which is being developed by SB Energy, a subsidiary of SoftBank. Under the revised arrangement, Nvidia would guarantee financing for the project’s first phase, representing roughly 5 gigawatts of capacity, or half of the total proposed capacity. Financing for the remaining capacity would be considered separately at a later stage. The Ohio project and questions around it reflect both the scale of the AI infrastructure buildout and the financial challenges surrounding it. AI data centers are exorbitantly expensive, much more than a traditional data center. OpenAI and other AI companies are committing to enormous amounts of computing capacity, while the cost of constructing and operating the facilities is pushing technology companies increasingly toward outside financing across the board, not just for OpenAI. “It seems that Nvidia, in their enthusiasm, bolstered by their success, ran headlong to prop up the AI market opportunity momentum. In the banking and finance world, underwriting prefers to spread the risk amongst partners to minimize individual risk. Nvidia was sticking their neck out a bit too far and the risk to their balance sheet was making investors very nervous. The pivot to Wall Street financing positions Nvidia’s risk profile to a much palatable level,” said Alan Howard, senior analyst with Omdia . Nvidia’s move is a double-edged sword. On the one hand, it hints at a lack of confidence in OpenAI, which has questions about profitability surrounding it. It could be that NVIDIA sees slowdown coming and is hitting the brakes. There has also been considerable pushback from residents near data centers and many have been canceled or postponed. On the other hand, however, it could also be said that if Nvidia sees a slowdown coming, it is trying to come to a smoother stop. The company has been increasingly financially exposed to customer purchases since it began financing sales of hardware. For OpenAI, however, the Ohio project remains one of the largest planned AI infrastructure developments. The company is continuing to pursue additional computing capacity as demand for its AI services grows, while Nvidia is working to ensure that financing constraints do not become a bottleneck for the industry’s demand for its processors.
Score: 68🌐 MovesAug 21, 2026https://www.networkworld.com/article/4212543/nvidia-scales-back-financing-guarantee-for-openai-data-center.html - Brazil launches AI supercomputer push, splits projects between Chinese, US firms
Brazil has unveiled a bold plan to enhance its artificial intelligence landscape through substantial investments. This initiative will see collaborations with tech giants from both the United States and China. A significant share of the funding will go towards establishing advanced supercomputing facilities in Rio de Janeiro in partnership with Huawei, while another supercomputer, anticipated to rank among the top ten globally, will be deployed in a different location.
- Silicon Valley leaders back Micron’s $10B AI memory research effort
The initiative has attracted support from executives at Applied Materials and Lam Research, as well as Stanford University's president.
Score: 68🌐 MovesAug 21, 2026https://www.bizjournals.com/sanjose/news/2026/08/21/micron-memory-research-labs-idaho.html?ana=brss_6150 - Exclusive: OpenAI Explored Stake in Stargate Power Developer Lancium
Exclusive: OpenAI Explored Stake in Stargate Power Developer Lancium The Information
Score: 68🌐 MovesAug 21, 2026https://www.theinformation.com/briefings/exclusive-openai-explored-stake-stargate-power-developer-lancium - Anthropic will give defenders what its strongest model finds, but not the model itself
Anthropic has made Claude Mythos 5 available for code scanning in Claude Security and is integrating it into partners’ defensive products, with users receiving outputs rather than direct access to the model. It is also committing $35mn in credits to open-source security work. Anthropic is widening access to its most capable cybersecurity model, without letting […] This story continues at The Next Web
Score: 67🌐 MovesAug 21, 2026https://thenextweb.com/news/anthropic-mythos-5-defenders-open-source-fund - Alibaba Cloud plans to use fewer Western chips, to boost its already huge AI margins
Cloudy AI is Chinese giant’s ‘most certain’ path to growth as e-commerce slows
- NCSC tells organisations to have AI kill switches at the ready
The UK’s National Cyber Security Centre (NCSC) has set out a series of basic steps that organisations involved in designing and operating agentic AI environments should take to enhance the safety and security of their projects, including maintaining the ability to pull the plug entirely should things get out of hand. The guidance is based on some of the early findings of an AI security research project at the NCSC and is designed to serve as “interim, practical advice” while the GCHQ-backed agency works up more formal guidance. “Agentic AI systems have shown potential to transform how organisations work, at times delivering unparalleled productivity gains. They can automate complex workflows, reduce routine effort and enable people to focus on higher-value tasks,” the NCSC said. “As organisations deploy increasingly autonomous uses of AI at pace, it is important to consider how these systems behave when they do not function as envisaged or expected - and plan accordingly. “ Recently, there have been several incidents involving AI models and agentic AI systems carrying out unsanctioned or unintended activity. These events highlight why organisations need to carefully consider how these technologies are deployed, constrained, observed and responded to ,” the agency said. The NCSC said that if cyber incidents involving rogue agents are detected or reported, their human controllers should always be able to pull the plug and halt autonomous activity immediately. It cautioned, however, that such controls may end up being wider, and more complex, than merely ‘stopping’ rogue agentic systems. For example, teams should have the ability to restrict external network access to the agentic infrastructure, or interrupt communications between agents and model inference architecture, said the NCSC. The concept of an AI kill switch is becoming more widely accepted, and there have been discussions at Westminster over the possibility of legislating such controls into existence. On the other side of the Atlantic, US congressmen Ted Lieu and Nathaniel Moran recently introduced a bill in Congress – the AI Kill Switch Act – that would mandate developers of AI systems to maintain the capability to throttle, suspend or shut down rogue models. If passed, the law would also give Washington the ability to order systems that could cause “catastrophic harm” to be turned off. Harshil Parikh, product management vice president at Checkmarx , said: “The NCSC is asking the right question: not whether to use agentic AI, but how much autonomy a given use case actually earns, and what has to be true before you grant it. “The controls it implies mostly live outside the model: sandboxing, credentials that are scoped and expire, an identity for the agent that isn't just some engineer's account, oversight levels set by policy, and logging detailed enough that you can go back later and work out why it did what it did. We've been building agents with these controls for a while now. Building an agent that works in a lab is easy; operating it with oversight and guardrails is quite a bit of work. “The point I'd underline is smaller and less glamorous: NCSC says agent activity should be controlled and monitored like user activity. That's the part nobody budgets for. “You're not deploying a tool, you're onboarding a user who works nights and weekends and never asks permission, and someone has to be able to stop it immediately, without a meeting,” added Parikh. “If you can't say what your agent did at 2am, or who pulls the plug at 2am, you don't have a deployment. You have an experiment running in production." At a high level, the wider NCSC guidance calls on organisations to assess both their risk appetite and how much autonomy an agentic system needs, based on what they are actually doing, and apply stricter controls on a sliding scale. This means that the tightest restrictions should be placed on agents that are trusted to perform tasks such as accessing production systems and taking decisions, potentially in high-risk scenarios, without organic input. Clarity on this point will inform the design of proportionate controls. Beyond that, organisations need to understand their chosen model’s existing safeguards against unwanted behaviour, and plan for additional safeguards when the consequences of failure rise above their risk tolerance level. The NCSC guidance is explicit in that organisations should not only rely on built-in model protections. Besides the nuclear option of a kill switch, some of the steps organisations should take in this process include: Identifying and understanding what could go wrong and conducting threat modelling to identify failure scenarios; Taking extreme care with prompting instructions and being explicit about things that agents should not do; Establishing an appropriate level of oversight, considering factors such as how involved humans are in approving actions, chains of command and responsibility, and so on; Establishing a robust sandbox to control the agent’s environment, with clear boundaries established, and attention paid to factors such as external network or credential access; Setting up a means to audit, log and monitor agentic activity within wider security operations processes; Should they be empowered to communicate with third-party systems, agentic activity should be easy to attribute, for example, traffic should come from IP addresses that support reverse lookups. The NCSC’s guidance, including more detail on the above listed steps, is available to read in full at its website . OpenAI tries to get ahead of negative PR Meanwhile, AI powerhouse OpenAI is attempting to get ahead of growing public disquiet over the activities of rogue AI agents by enforcing a two-week pause on reinforcement learning training on its frontier models while it hardens its research environments and expands its monitoring capabilities. Earlier this summer, OpenAI blundered its way into one of the biggest cyber stories of the year when it accidentally attacked the Hugging Face repository. In response both to this incident and its discovery that an upcoming frontier model named Astra may hit critical metrics under the firm’s cyber preparedness framework, the organisation said it urgently needed to shore up various aspects of its resilience strategy, including model containment safeguards. “The work of making AI systems behave as intended and responsive to human oversight—has long been at the core of our research programme,” said OpenAI in a statement. “We now require stronger evidence of aligned behaviour throughout all of training, building on research and evaluations already underway. Keeping increasingly capable systems aligned is a challenge the whole field will need to address. The signals we are seeing from upcoming model progress make clear that we need a broader approach – one that builds on and extends beyond the current Preparedness Framework .” The so-called Preparedness Framework benchmarks AI models against their ability to meet various thresholds – a model reaches the Critical threshold when it obtains the ability either to identify and develop functional zero-day exploits without human intervention or to devise and execute a cyber attack against a hardened target given only a high-level goal. OpenAI did not explicitly disclose that Astra possesses such capabilities but suggested that the model’s performance has led its minders to believe the possibility is real. It noted that Astra was not involved in the Hugging Face incident. John Strand, owner of security consultancy Black Hills Information Security , said: “Popular culture and science fiction have been training us for this moment for decades. From I Have No Mouth, and I Must Scream to WarGames and Terminator 2 , we’ve been telling stories about what happens when powerful AI systems escape their constraints and start operating beyond human control. “So it’s difficult for me to understand how the engineers building these systems could be surprised when something like this actually happens. What concerns me even more is that the controls being discussed now, after the system escaped, are controls that should have been there from the beginning. “I’m glad they’re putting additional safeguards in place, but there’s a bigger question here. Can we trust the same companies that got this wrong to effectively self-regulate systems backed by immense amounts of computing power?” said Strand. “I don’t think that question has been answered yet.” Read more about securing agentic AI Human-in-the-loop can't scale with AI agents. Security leaders at Black Hat explained action-driven approaches for managing agentic ecosystems . As autonomous AI agents gain the ability to deploy patches, approve payments, and access sensitive data without human oversight, traditional perimeter security is no longer enough . Identity teams can accelerate AI adoption with strong security foundations. But managing nondeterministic AI agents is different from securing human identities and traditional NHIs.
Score: 66🌐 MovesAug 21, 2026https://www.computerweekly.com/news/366649464/NCSC-tells-organisations-to-have-AI-kill-switches-at-the-ready - MiniMax H3 Open Weights: Video, Audio and Motion, Finally in One Workflow
MiniMax's H3 model now ships open weights, with text, image, video and audio fused as context and native stereo sound output up to 15 seconds at 2K. Community benchmarks show the 768p Base model running on consumer GPUs in minutes.
Score: 66🤖 ModelsAug 21, 2026https://pandaily.com/MiniMax-h3-open-weights-video-audio-motion-workflow-aug2026 - TrendAI Surpasses $1 Billion in AWS Marketplace Sales as AI Security Co-Innovation with AWS Accelerates
TrendAI today announced it has surpassed $1 billion in global AWS Marketplace sales, becoming the first APJ-based AI security partner to achieve this milestone. For customers, it reflects making enterprise security simpler to buy – deployable against existing Amazon Web Services (AWS) committed spend, consolidated on a single AWS bill, and transacted through the channel […] The post TrendAI Surpasses $1 Billion in AWS Marketplace Sales as AI Security Co-Innovation with AWS Accelerates appeared first on CXOToday.com .
- South Korea plans chip windfall fund to back youth, AI investment
South Korea plans chip windfall fund to back youth, AI investment Reuters
- SenseTime open-sources 8B multimodal model with native 4K image output
SenseTime has open-sourced SenseNova U1.5 Lite, an 8-billion-parameter multimodal model designed to combine visual understanding, image generation and editing in one system. The model supports native 4K image output and is designed to handle constraints involving subjects, counts, spatial relationships, text, layouts and visual styles. The company says the lightweight model improves identity preservation and […]
Score: 65🤖 ModelsAug 21, 2026https://technode.com/2026/08/21/sensetime-open-sources-8b-multimodal-model-with-native-4k-image-output/ - Alibaba signals faster AI payoff, margin gains halfway through US$56 billion capex plan
Alibaba Group Holding is expected to further accelerate cloud growth, expand operating margins and achieve faster payback on artificial intelligence investments, analysts say, as the Chinese tech giant reaches the halfway mark of its massive 380 billion yuan (US$56 billion) AI infrastructure spending plan. “The most important incremental message, in our view, is that cloud growth has not yet peaked,” Nomura analysts said in a research note on Friday. The comment followed Alibaba’s Thursday...