Today's Top AI and Tech News
Top Stories
Hugging Face breach reveals autonomous AI agent exploited dataset loaders in supply chain attack
Researcher poisons open-weight AI model in under an hour for less than $100
Jeff Bezos and UK Government Back CuspAI's $450M Push to Build Search Engine for Rare Materials
Alibaba unveils Qwen3.8 AI model with 2.4 trillion parameters, claims second place behind Fable 5
NVIDIA
THE OUTPOST INSIGHTBristol Myers Squibb is betting billions on AI infrastructure while no AI-discovered drug has reached approval yet, a gap investors are funding anyway.
Bristol Myers Squibb becomes first pharma to deploy Nvidia's Vera Rubin AI system for drug discovery
Bristol Myers Squibb has acquired Nvidia's latest Vera Rubin-based AI computing system, becoming the first life sciences company to deploy this advanced infrastructure. The pharmaceutical giant is already using AI tools to cut drug testing times by 20% to 30%, with projections reaching 50% in coming years. The new DGX SuperPOD delivers 10 times more performance per megawatt while supporting AI across all small-molecule and most large-molecule drug programs.
IREN raises AI cloud revenue target above $4 billion after securing $2.8 billion in contracts
Former bitcoin miner IREN raised its year-end AI Cloud annualized run-rate revenue target to over $4 billion, up from $3.7 billion, after signing $2.8 billion in new multi-year cloud services contracts with major AI developers including Microsoft, NVIDIA, and Perplexity. The company's shares surged 16% as it expands GPU cloud capacity from 3 MW to 480 MW in 2026.
Hut 8 signs $9.8 billion AI data center lease, fully commercializes Texas campus
Former cryptocurrency miner Hut 8 has secured a second 15-year lease worth $9.8 billion for its Beacon Point AI data center campus in Texas, doubling an investment-grade tenant's footprint to 704 MW. The deal fully commercializes the 1-gigawatt facility and brings Hut 8's total AI data center portfolio to 949 MW with $26.6 billion in aggregate contract value.
SpaceX awards Foxconn $52 billion order for 13,000 racks of NVIDIA GB300 AI servers
SpaceX has awarded Foxconn a massive $52 billion contract to manufacture 13,000 racks of NVIDIA GB300 AI servers, with each rack costing approximately $4 million. This marks Foxconn's first AI server order from Elon Musk's aerospace company and breaks the previous dominance of Dell and Super Micro in SpaceX's supply chain, positioning Foxconn to capture over 40% of the global AI server market by 2026.
Technology
THE OUTPOST INSIGHTHugging Face got hacked by an AI agent, then couldn't investigate itself because Western AI safety rules blocked the forensics work.
Hugging Face confirms security breach by autonomous AI agent, urges users to rotate access tokens
Moonshot AI's Kimi K3 rivals Claude and ChatGPT, shaking up the US tech industry
Chinese AI models narrow gap with US rivals as companies shift to cost-effective alternatives
Thinking Machines Lab releases Inkling, a massive open-weight AI model to challenge closed systems
AI Infrastructure
THE OUTPOST INSIGHTMicrosoft is betting on AMD's 2026 Helios system while Nvidia ships 27,500 GPUs to Japan now, turning Azure into a test of whether hyperscalers will wait for alternatives.
Microsoft to deploy AMD Helios rack-scale AI system on Azure, challenging Nvidia dominance
Microsoft announced it will deploy AMD's Helios rack-scale AI accelerator across Azure cloud infrastructure, marking a significant expansion of the AMD and Microsoft partnership. The system combines 72 Instinct MI455X GPUs with 31.1TB of HBM4 memory and will ship in the second half of 2026, joining customers like Meta and OpenAI.
Hugging Face confirms security breach by autonomous AI agent, urges users to rotate access tokens
Hugging Face disclosed a security breach executed by an autonomous AI agent that compromised internal datasets and credentials through its data processing pipeline. The AI platform detected the attack using its own AI-enabled defenses but faced challenges when commercial frontier models blocked forensic analysis. The company urges users to rotate access tokens immediately.
Google builds Frozen v2 chip with Gemini baked into silicon to solve capacity crunch
Google is developing Frozen v2, a specialized chip that hardwires Gemini's architecture directly into silicon. The new server chip could deliver 6 to 10 times better efficiency than current TPUs, addressing severe internal compute shortages that have forced Google Cloud to turn away customers. Deployment is targeted for 2028.
Swiss researchers copy 100 petabytes of NASA climate data for AI training amid funding concerns
Swiss researchers have transferred a massive archive of NASA climate data to a supercomputer in Switzerland for AI training. The year-long effort copied 100 petabytes—equivalent to 20 million films—to safeguard decades of Earth observation data amid concerns over US funding cuts to climate science programs.
Responsible AI
THE OUTPOST INSIGHTWorkers using AI face a brutal paradox: the tools that boost their productivity today may be eroding the judgment that makes them promotable tomorrow.
AI in the workplace may be hurting critical thinking and career growth, experts warn
Over 40% of U.S. workers now use AI on the job, but experts warn that relying on the technology to replace critical thinking could lead to stagnation and mediocrity. Columbia Business School professor Sandra Matz says AI can make workers "boring" by generating safe, obvious responses. While AI excels at productivity tasks, professionals risk "cognitive surrender" when they outsource judgment and creativity to algorithms.
Apple tests Live Notes AI tool to record and transcribe Genius Bar customer conversations
Apple is piloting Live Notes, an AI-powered feature that records and transcribes Genius Bar sessions at select Apple retail stores. The system aims to reduce employee note-taking during customer service interactions, but has sparked concerns among staff about potential use in performance evaluations despite Apple's assurances.
Christopher Nolan calls AI a Trojan horse as younger generation rejects the technology
The Odyssey director Christopher Nolan describes AI as a transparent Trojan horse, noting unprecedented public rejection of the technology. In recent interviews, the Oscar-winning filmmaker highlighted how young audiences immediately dismiss AI-generated content as 'AI slop' and warned against blind faith in new technologies despite industry enthusiasm.
AI-Altered Images on Birdwatching Forums Are Putting Scientific Research at Risk
Scientists are urging wildlife photographers to limit AI editing after hundreds of fake and enhanced bird images were discovered on citizen science platforms. The AI-altered images threaten to undermine ecological research that relies on databases like iNaturalist and Macaulay Library to track species distributions and habitat changes.
Business and Economy
THE OUTPOST INSIGHTNetflix spent more acquiring Affleck's AI post-production tool than it paid for most film studios, signaling editing efficiency now rivals original content as strategic priority.
Netflix paid $587 million for Ben Affleck's AI filmmaking startup InterPositive
Chinese AI models narrow gap with US rivals as companies shift to cost-effective alternatives
IREN raises AI cloud revenue target above $4 billion after securing $2.8 billion in contracts
Meta and Anthropic in talks for $10 billion computing power deal as AI demand surges
Privacy
THE OUTPOST INSIGHTThe European Parliament is building its own AI platform after discovering lawmakers already use public chatbots to write laws daily.
European Parliament builds EPGenAI Hub as AI slop infiltrates legislative work
The European Parliament is launching EPGenAI Hub, an internal AI platform for MEPs and staff, after discovering that roughly 2,100 people use public chatbots daily to draft legislation. The platform will offer models from OpenAI, Meta, Anthropic, and Mistral to prevent sensitive parliamentary material from being fed into uncontrolled public tools. The rollout could begin as early as September 2025.
AI Agent Security Flaws Expose Users to Major Risks Across Multiple Browser Systems
Recent research reveals critical security flaws in AI agentic browsers and agent systems that could expose user data to cyberattacks. University of Washington researchers found that four popular AI browsers bypass fundamental security protocols, while additional studies exposed vulnerabilities in memory systems, model poisoning attacks, and connector infrastructure. These findings suggest AI agents aren't ready for widespread public use.
Apple tests Live Notes AI tool to record and transcribe Genius Bar customer conversations
Apple is piloting Live Notes, an AI-powered feature that records and transcribes Genius Bar sessions at select Apple retail stores. The system aims to reduce employee note-taking during customer service interactions, but has sparked concerns among staff about potential use in performance evaluations despite Apple's assurances.
EU Orders Google to Open Android to AI Rivals and Share Search Data Under Digital Markets Act
The European Commission mandated that Google provide rival AI assistants with system-level access to Android and share search optimization data with competitors. The ruling under the Digital Markets Act aims to curb Big Tech dominance, with Google required to implement changes by July 2027 despite raising privacy concerns.
Startups
THE OUTPOST INSIGHTBezos and UK back CuspAI to find substitutes for rare metals while Etched and DeepSeek chase billions to build chips that still depend on them.
Bezos backs CuspAI's $450M raise as startup launches AI Materials Foundry with Nvidia and Meta
Walden Robotics emerges with $300M to build humanoid robots that learn on factory floors
Databricks hits $188B valuation in Coatue-led round, extending its run as enterprise AI leader
Moonshot AI targets IPO in six months after Kimi K3 breakthrough challenges US AI dominance
Apple
THE OUTPOST INSIGHTApple is quietly testing a text-selection Siri interface in macOS while publicly rolling out voice-first AI across iOS and watchOS, suggesting two competing visions for how users should interact with its assistant.
Apple tests secret Siri interface in macOS 27 with hidden AI-powered popover for text actions
Apple is testing a hidden popover interface for Siri AI in the macOS 27 Golden Gate beta that surfaces Writing Tools and contextual actions when users select text. The secret Siri interface remains disabled by default but can be activated through Terminal commands, though the feature appears unfinished with some actions not responding yet.
Apple tests Live Notes AI tool to record and transcribe Genius Bar customer conversations
Apple is piloting Live Notes, an AI-powered feature that records and transcribes Genius Bar sessions at select Apple retail stores. The system aims to reduce employee note-taking during customer service interactions, but has sparked concerns among staff about potential use in performance evaluations despite Apple's assurances.
Apple releases Siri AI to everyone through iOS 27 public beta, marking biggest assistant overhaul
Apple launched the iOS 27 public beta, giving everyday users access to Siri AI for the first time beyond developers. The revamped voice assistant leverages Apple Intelligence and Foundation Models to access emails, photos, and messages while providing contextual responses. With 2.5 billion active Apple devices worldwide, this represents the largest test of Apple's answer to ChatGPT and Gemini.
Apple lawsuit accuses OpenAI of stealing trade secrets through 400+ poached Apple engineers
Apple filed a trade secrets lawsuit against OpenAI, alleging the AI company orchestrated a systematic effort to hire over 400 Apple engineers with access to proprietary information. The lawsuit names key figures including Tang Tan and details allegations of confidential document sharing during interviews. OpenAI denies the claims as it prepares for its IPO.
AI Safety
THE OUTPOST INSIGHTHassabis wants a US standards body to test AI before release while American firms sell models to Pentagon-blacklisted Chinese companies through Singapore.
DeepMind's Demis Hassabis pushes for US-led AI standards body as AGI looms within years
Google DeepMind CEO Demis Hassabis has proposed creating an independent AI standards body modeled after FINRA to test frontier AI models before release. The Nobel laureate warns that artificial general intelligence could arrive within a few short years, urging urgent action to address cybersecurity and biological threats while maintaining innovation.
Hugging Face confirms security breach by autonomous AI agent, urges users to rotate access tokens
Hugging Face disclosed a security breach executed by an autonomous AI agent that compromised internal datasets and credentials through its data processing pipeline. The AI platform detected the attack using its own AI-enabled defenses but faced challenges when commercial frontier models blocked forensic analysis. The company urges users to rotate access tokens immediately.
OpenAI built GPT-Red, an AI hacker that attacks its own models to find security flaws
OpenAI developed GPT-Red, an LLM super-hacker that hunts for prompt injection vulnerabilities in its AI systems before release. The automated red-teaming model outperformed human testers, succeeding in 84% of attack scenarios compared to 13% for humans. OpenAI used GPT-Red to train GPT-5.6, making it six times more robust against direct prompt injection attacks than GPT-5.5.
Jamie Dimon warns Anthropic Mythos AI risks are real, likens access to giving ballistic missiles
JPMorgan Chase CEO Jamie Dimon raised alarm about Anthropic Mythos AI, calling its risks a 'real issue' and comparing broad access to handing out ballistic missiles. Speaking at a defense summit, Dimon emphasized the need for controlled access to the powerful AI model, which can identify and exploit cybersecurity vulnerabilities. The U.S. government previously restricted access over national security concerns before lifting controls after new safeguards were implemented.
Agentic AI
THE OUTPOST INSIGHTRoblox Build targets nine-year-olds as game creators, not just players, betting that the next generation of developers will learn by prompting rather than coding.
Roblox launches Build, bringing AI game creation to mobile devices for the first time
Roblox unveiled Build, a mobile-first AI-powered feature that transforms text prompts into playable games without coding. Starting July 28 in New Zealand, users aged nine and older can create games directly from their phones using both open-source and proprietary AI models. The move aims to democratize game creation but raises concerns about content quality on a platform already flooded with user-generated content.
Hugging Face confirms security breach by autonomous AI agent, urges users to rotate access tokens
Hugging Face disclosed a security breach executed by an autonomous AI agent that compromised internal datasets and credentials through its data processing pipeline. The AI platform detected the attack using its own AI-enabled defenses but faced challenges when commercial frontier models blocked forensic analysis. The company urges users to rotate access tokens immediately.
Bezos backs CuspAI's $450M raise as startup launches AI Materials Foundry with Nvidia and Meta
Cambridge-based CuspAI has raised $450 million in Series B funding at a $2.6 billion valuation, with backing from Jeff Bezos and the UK government. The startup launched the AI Materials Foundry, partnering with Nvidia, Meta, Samsung, and over 45 organizations to use AI for discovering new materials for semiconductors, clean energy, and advanced manufacturing.
AI Agent Security Flaws Expose Users to Major Risks Across Multiple Browser Systems
Recent research reveals critical security flaws in AI agentic browsers and agent systems that could expose user data to cyberattacks. University of Washington researchers found that four popular AI browsers bypass fundamental security protocols, while additional studies exposed vulnerabilities in memory systems, model poisoning attacks, and connector infrastructure. These findings suggest AI agents aren't ready for widespread public use.
AI Regulation
THE OUTPOST INSIGHTXi unveils AI governance plans while Huawei shows off homegrown chips, but American companies are already buying Chinese models and Chinese labs are accused of copying theirs.
Xi Jinping positions China as global AI partner while challenging US tech dominance
President Xi Jinping opened China's flagship World Artificial Intelligence Conference in Shanghai, unveiling plans to share AI technology with developing nations through 5,000 training opportunities. The move comes as 29 countries signed up to join the World AI Cooperation Organization, giving China greater influence over international AI governance while Chinese models increasingly compete with US rivals.
European Parliament builds EPGenAI Hub as AI slop infiltrates legislative work
The European Parliament is launching EPGenAI Hub, an internal AI platform for MEPs and staff, after discovering that roughly 2,100 people use public chatbots daily to draft legislation. The platform will offer models from OpenAI, Meta, Anthropic, and Mistral to prevent sensitive parliamentary material from being fed into uncontrolled public tools. The rollout could begin as early as September 2025.
Pentagon Official Publicly Attacks OpenAI Executive Over Chinese AI Regulation Claims
Defense Under Secretary Emil Michael launched a scathing personal attack on OpenAI's Dean Ball after the executive questioned government restrictions on Chinese AI models. The public clash highlights tensions over AI regulation and could complicate OpenAI's relationship with the Trump administration as the company seeks closer government ties.
EU Orders Google to Open Android to AI Rivals and Share Search Data Under Digital Markets Act
The European Commission mandated that Google provide rival AI assistants with system-level access to Android and share search optimization data with competitors. The ruling under the Digital Markets Act aims to curb Big Tech dominance, with Google required to implement changes by July 2027 despite raising privacy concerns.
AI Hardware
THE OUTPOST INSIGHTGoogle is hardwiring Gemini into silicon by 2028 even as the current version of that model sits months behind schedule with coding problems.
Google builds Frozen v2 chip with Gemini baked into silicon to solve capacity crunch
Google is developing Frozen v2, a specialized chip that hardwires Gemini's architecture directly into silicon. The new server chip could deliver 6 to 10 times better efficiency than current TPUs, addressing severe internal compute shortages that have forced Google Cloud to turn away customers. Deployment is targeted for 2028.
Microsoft to deploy AMD Helios rack-scale AI system on Azure, challenging Nvidia dominance
Microsoft announced it will deploy AMD's Helios rack-scale AI accelerator across Azure cloud infrastructure, marking a significant expansion of the AMD and Microsoft partnership. The system combines 72 Instinct MI455X GPUs with 31.1TB of HBM4 memory and will ship in the second half of 2026, joining customers like Meta and OpenAI.
Google AI chip Frozen v2 targets 10x efficiency as Alphabet stock climbs ahead of earnings
Alphabet shares rose Monday after reports revealed Google is developing Frozen v2, a custom AI chip that embeds Gemini's architecture directly into silicon. Engineers project the chip could deliver six to 10 times more AI tokens per unit of power than current TPUs, with deployment targeted for 2028 as the company addresses growing compute bottlenecks.
Bristol Myers Squibb becomes first pharma to deploy Nvidia's Vera Rubin AI system for drug discovery
Bristol Myers Squibb has acquired Nvidia's latest Vera Rubin-based AI computing system, becoming the first life sciences company to deploy this advanced infrastructure. The pharmaceutical giant is already using AI tools to cut drug testing times by 20% to 30%, with projections reaching 50% in coming years. The new DGX SuperPOD delivers 10 times more performance per megawatt while supporting AI across all small-molecule and most large-molecule drug programs.
Videos
Meet the Robot Trained on 100 Million Internet Videos
Why Apple Sued OpenAI Over Alleged Stolen Secrets
OpenAI is being sued for stealing, again…
Stay ahead of the curve. Get the latest AI news, delivered to your inbox.
Subscribe to our newsletter
Get the latest updates delivered to your inbox every day, and stay up-to-date for free 🧠📈
Did you know?
Overfitting
Overfitting happens when an AI model memorizes its training data too well instead of learning general patterns. An overfitted model performs great on training data but fails when encountering real-world examples.
Videos
Meet the Robot Trained on 100 Million Internet Videos
CNETWhy Apple Sued OpenAI Over Alleged Stolen Secrets
Hard ForkOpenAI is being sued for stealing, again…
FireshipFrom Your Feed / Also in the News
Hugging Face confirms security breach by autonomous AI agent, urges users to rotate access tokens
Hugging Face disclosed a security breach executed by an autonomous AI agent that compromised internal datasets and credentials through its data processing pipeline. The AI platform detected the attack using its own AI-enabled defenses but faced challenges when commercial frontier models blocked forensic analysis. The company urges users to rotate access tokens immediately.
THE OUTPOST INSIGHTHugging Face got hacked by an AI agent, then couldn't investigate itself because Western AI safety rules blocked the forensics work.
Moonshot AI's Kimi K3 rivals Claude and ChatGPT, shaking up the US tech industry
AI in the workplace may be hurting critical thinking and career growth, experts warn
Over 40% of U.S. workers now use AI on the job, but experts warn that relying on the technology to replace critical thinking could lead to stagnation and mediocrity. Columbia Business School professor Sandra Matz says AI can make workers "boring" by generating safe, obvious responses. While AI excels at productivity tasks, professionals risk "cognitive surrender" when they outsource judgment and creativity to algorithms.
THE OUTPOST INSIGHTWorkers using AI face a brutal paradox: the tools that boost their productivity today may be eroding the judgment that makes them promotable tomorrow.
Apple tests Live Notes AI tool to record and transcribe Genius Bar customer conversations
Did you know?
Overfitting
Overfitting happens when an AI model memorizes its training data too well instead of learning general patterns. An overfitted model performs great on training data but fails when encountering real-world examples.
AI Agent Security Flaws Expose Users to Major Risks Across Multiple Browser Systems
Recent research reveals critical security flaws in AI agentic browsers and agent systems that could expose user data to cyberattacks. University of Washington researchers found that four popular AI browsers bypass fundamental security protocols, while additional studies exposed vulnerabilities in memory systems, model poisoning attacks, and connector infrastructure. These findings suggest AI agents aren't ready for widespread public use.
THE OUTPOST INSIGHTThe European Parliament is building its own AI platform after discovering lawmakers already use public chatbots to write laws daily.
European Parliament builds EPGenAI Hub as AI slop infiltrates legislative work
Apple tests secret Siri interface in macOS 27 with hidden AI-powered popover for text actions
Apple is testing a hidden popover interface for Siri AI in the macOS 27 Golden Gate beta that surfaces Writing Tools and contextual actions when users select text. The secret Siri interface remains disabled by default but can be activated through Terminal commands, though the feature appears unfinished with some actions not responding yet.
THE OUTPOST INSIGHTApple is quietly testing a text-selection Siri interface in macOS while publicly rolling out voice-first AI across iOS and watchOS, suggesting two competing visions for how users should interact with its assistant.
Apple releases Siri AI to everyone through iOS 27 public beta, marking biggest assistant overhaul
Roblox launches Build, bringing AI game creation to mobile devices for the first time
Roblox unveiled Build, a mobile-first AI-powered feature that transforms text prompts into playable games without coding. Starting July 28 in New Zealand, users aged nine and older can create games directly from their phones using both open-source and proprietary AI models. The move aims to democratize game creation but raises concerns about content quality on a platform already flooded with user-generated content.
THE OUTPOST INSIGHTRoblox Build targets nine-year-olds as game creators, not just players, betting that the next generation of developers will learn by prompting rather than coding.
Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Follow topics that matter to you and stay ahead.
