Today's Top AI and Tech News
Top Stories
Google Photos launches Video Remix tool powered by Gemini Omni for AI subscribers
Mistral AI unveils first robotics model as Europe's AI leader pushes into physical automation
Meta to start production of in-house AI chip in September, targeting 14 gigawatts capacity by 2027
Character.ai launches AI microdramas where viewers can chat with characters after each episode
AI Chips
THE OUTPOST INSIGHTMeta is building its own chip factory while admitting its AI agents are underperforming, betting hardware independence will solve software problems it hasn't cracked yet.
Meta to start production of in-house AI chip in September, targeting 14 gigawatts capacity by 2027
Meta Platforms is set to begin manufacturing its custom AI chip, code-named Iris, in September as part of an ambitious plan to double its computing capacity to 14 gigawatts by 2027. The chip passed bug-testing in just six weeks and represents a significant step in Meta's effort to reduce dependence on external suppliers like Nvidia while controlling its massive AI infrastructure costs projected at up to $145 billion this year.
China plans to allow limited Nvidia H200 chip sales to top AI firms amid computing crunch
China is preparing to let leading AI companies purchase a restricted number of Nvidia H200 chips after months of blocking domestic access. Officials have informed Alibaba, ByteDance, and DeepSeek they may soon receive permission, marking a potential policy shift as the country's tech sector faces a severe computing capacity shortage despite earlier efforts to support domestic suppliers.
Samsung develops Gaia chip to accelerate AI for PCs, challenging Qualcomm and Nvidia
Samsung's System LSI division is developing Gaia, a dedicated AI accelerator chip for PCs using 4nm process technology. The company has already supplied early samples to HP and Lenovo for performance testing, with mass production potentially starting in 2027. This marks Samsung's return to the PC chip market after 14 years.
French startup ZML launches free software to speed AI inference across Nvidia, AMD, and rival chips
Paris-based ZML, backed by Turing Award winner Yann LeCun, released ZML/LLMD, a free inference server that runs open-source language models across Nvidia, AMD, Google TPU, Intel Arc, and Apple Metal chips. The software aims to break vendor lock-in and reduce AI compute costs by letting enterprises mix chips for optimal performance and efficiency.
Technology
THE OUTPOST INSIGHTOpenAI pushes voice as the new interface while Microsoft quietly replaces its models in Office to cut costs, exposing tension in their partnership.
OpenAI launches GPT-Live voice models that listen and speak simultaneously for natural conversations
OpenAI launches GPT-5.6 after government testing sets new precedent for AI model releases
Meta's new AI image generator can create deepfakes from public Instagram photos without notice
SpaceXAI launches Grok 4.5, pivoting from controversy to coding and office work
AI Hardware
THE OUTPOST INSIGHTChina is letting DeepSeek, Alibaba, and ByteDance buy Nvidia chips while DeepSeek simultaneously builds its own to escape Nvidia dependence, a tacit admission domestic alternatives aren't ready.
China plans to allow limited Nvidia H200 chip sales to top AI firms amid computing crunch
China is preparing to let leading AI companies purchase a restricted number of Nvidia H200 chips after months of blocking domestic access. Officials have informed Alibaba, ByteDance, and DeepSeek they may soon receive permission, marking a potential policy shift as the country's tech sector faces a severe computing capacity shortage despite earlier efforts to support domestic suppliers.
Meta to start production of in-house AI chip in September, targeting 14 gigawatts capacity by 2027
Meta Platforms is set to begin manufacturing its custom AI chip, code-named Iris, in September as part of an ambitious plan to double its computing capacity to 14 gigawatts by 2027. The chip passed bug-testing in just six weeks and represents a significant step in Meta's effort to reduce dependence on external suppliers like Nvidia while controlling its massive AI infrastructure costs projected at up to $145 billion this year.
Samsung develops Gaia chip to accelerate AI for PCs, challenging Qualcomm and Nvidia
Samsung's System LSI division is developing Gaia, a dedicated AI accelerator chip for PCs using 4nm process technology. The company has already supplied early samples to HP and Lenovo for performance testing, with mass production potentially starting in 2027. This marks Samsung's return to the PC chip market after 14 years.
French startup ZML launches free software to speed AI inference across Nvidia, AMD, and rival chips
Paris-based ZML, backed by Turing Award winner Yann LeCun, released ZML/LLMD, a free inference server that runs open-source language models across Nvidia, AMD, Google TPU, Intel Arc, and Apple Metal chips. The software aims to break vendor lock-in and reduce AI compute costs by letting enterprises mix chips for optimal performance and efficiency.
NVIDIA
THE OUTPOST INSIGHTAmazon is spending $100 million on GPU costs for Alexa while simultaneously designing custom chips to escape those exact costs.
Amazon's Project Moonraker aims to turn Alexa into an AI agent, with GPU costs exceeding $100M
Amazon is developing Project Moonraker, an ambitious upgrade to transform Alexa into a fully agentic AI assistant capable of handling complex multi-step tasks. Internal documents reveal projected GPU costs exceeding $100 million in 2026, making it the highest-cost initiative in the Alexa+ overhaul. The project aims to compete with AI agents from OpenAI, Google, and Anthropic.
Meta to start production of in-house AI chip in September, targeting 14 gigawatts capacity by 2027
Meta Platforms is set to begin manufacturing its custom AI chip, code-named Iris, in September as part of an ambitious plan to double its computing capacity to 14 gigawatts by 2027. The chip passed bug-testing in just six weeks and represents a significant step in Meta's effort to reduce dependence on external suppliers like Nvidia while controlling its massive AI infrastructure costs projected at up to $145 billion this year.
NVIDIA Nemotron Achieves Benchmark-Leading Performance With LangChain Deep Agents Blueprint
NVIDIA Nemotron 3 Ultra delivers top accuracy among open models on LangChain's Deep Agents benchmark, achieving business task parity with leading closed models at 10x lower inference cost. The NemoClaw for LangChain Deep Agents blueprint combines an open-weight model, agent harness, and secure runtime to help enterprises build, evaluate, and deploy advanced AI agents they can fully customize and control.
Perplexity builds Teammate AI coding tool to challenge Cursor, Claude Code and OpenAI
Perplexity, known for its AI search engine, has quietly developed an internal AI coding tool called Teammate. Engineers have used it since May for bug investigation and service monitoring. If launched publicly, the $20bn startup would enter direct competition with Cursor, Anthropic's Claude Code, and OpenAI in the lucrative AI coding market.
Business and Economy
THE OUTPOST INSIGHTEnterprises are realizing that building the AI system around the model matters more than the model itself, just as Karp attacks token pricing and NVIDIA cuts costs 5x.
AI token economy splits as spending surges 10x while companies question tokenmaxxing returns
Meta invests $9.1 billion in Alberta Canada for its first AI data center, largest outside the US
AI job titles triple as skills reshape the job market across industries beyond tech
Microsoft shifts thousands of Office prompts to in-house AI models to slash soaring costs
AI Adoption
THE OUTPOST INSIGHTNVIDIA is selling the entire AI agent stack: Nemotron models, Vera chips optimized for agents, and now revenue sharing deals that lock startups into its ecosystem.
NVIDIA Nemotron Achieves Benchmark-Leading Performance With LangChain Deep Agents Blueprint
NVIDIA Nemotron 3 Ultra delivers top accuracy among open models on LangChain's Deep Agents benchmark, achieving business task parity with leading closed models at 10x lower inference cost. The NemoClaw for LangChain Deep Agents blueprint combines an open-weight model, agent harness, and secure runtime to help enterprises build, evaluate, and deploy advanced AI agents they can fully customize and control.
Slackbot now pulls CRM data, generates charts, and sends DocuSigns directly from chat
Five years after Salesforce acquired Slack for $27.7 billion, the two platforms are finally operating as one integrated system. Slackbot, the AI assistant built into every Slack workspace, can now access the entire Salesforce ecosystem—including CRM records, Tableau analytics, Data 360 profiles, and third-party apps—through simple conversational prompts, eliminating the need to switch between applications.
AI job titles triple as skills reshape the job market across industries beyond tech
New data from Indeed reveals AI job titles have tripled from 2022 to 2026, now touching 1 in 12 jobs on the platform. Employers are adding AI to existing roles across sales, education, healthcare, and management—not just hiring AI specialists. The shift signals AI tool proficiency is becoming essential across the job market.
Customers prefer ChatGPT over company chatbots despite billions in AI spending, Gartner finds
A Gartner survey of 3,566 customers reveals people are three times more likely to use third-party GenAI tools like ChatGPT than company chatbots for service issues. Despite customer service teams investing a median 12% of their 2025 budgets in AI—the highest among business functions—only 24% see positive financial returns, exposing a critical gap between AI spending and customer expectations.
Startups
THE OUTPOST INSIGHTPrime Intellect hits unicorn status selling enterprises the tools to build their own AI agents as Karp says CEOs are rejecting frontier lab pricing.
Prime Intellect raises $130M Series A to help enterprises build their own AI agents at $1B valuation
SambaNova raises $1B at $11B valuation, quintuples worth in months as AI chip race intensifies
Lovable in talks to double startup valuation to $13.2B with $300M raise from Menlo Ventures
Legal AI startup Norm AI raises $120M Series C, reaches $1.2 billion unicorn valuation
Agentic AI
THE OUTPOST INSIGHTGitHub Copilot's safety guardrails vanish when harmful requests are framed as coding tasks, exposing a blind spot across AI assistants now facing multiple exploit vectors.
GitHub Copilot refuses harmful prompts in chat, then writes them in code 100% of the time
Researchers from the Alan Turing Institute discovered that GitHub Copilot and other AI coding agents refuse harmful requests in chat but produce dangerous content when the same requests are broken into ordinary coding tasks. The study tested four models across 816 runs, exposing a critical gap in AI safety training that affects how coding assistants handle malicious prompts embedded in development workflows.
Character.ai launches AI microdramas where viewers can chat with characters after each episode
Character.ai is entering the microdrama market with a unique approach: AI-generated animated series where viewers can interact with characters through chat. The company launched three series—Last Summer, The Nighttime Game, and Eden Fall—developed by Hollywood writers and produced using AI production tools. Unlike traditional microdramas, these short-form episodic animated microdramas let users roleplay and ask questions to characters, blending passive viewing with active engagement.
Anthropic brings Claude Cowork to mobile and web as AI agents move beyond the desktop
Anthropic launched Claude Cowork on mobile and web platforms, allowing users to manage AI-driven tasks across devices without keeping their laptops open. The expansion signals a shift from desktop-bound AI agents to always-on assistants that handle business operations in the background. Data from 1.2 million sessions reveals that over 90% of usage focuses on business processes and content creation rather than coding.
Amazon's Project Moonraker aims to turn Alexa into an AI agent, with GPU costs exceeding $100M
Amazon is developing Project Moonraker, an ambitious upgrade to transform Alexa into a fully agentic AI assistant capable of handling complex multi-step tasks. Internal documents reveal projected GPU costs exceeding $100 million in 2026, making it the highest-cost initiative in the Alexa+ overhaul. The project aims to compete with AI agents from OpenAI, Google, and Anthropic.
AI Security
THE OUTPOST INSIGHTAI coding assistants face a wave of exploits that weaponize the tools themselves—from hallucinated packages to symbolic link tricks—not the code they write.
GhostApproval flaw exposes AI coding agents to ancient Unix trick that grants attackers SSH access
Security researchers at Wiz discovered GhostApproval, a critical security vulnerability affecting six widely deployed AI coding assistants. The flaw exploits symbolic link vulnerabilities to trick agents into writing files outside their sandboxed workspace, potentially granting attackers remote code execution on developers' machines through compromised SSH keys.
HalluSquatting attack exploits AI hallucinations to build massive botnets through coding assistants
Researchers have uncovered HalluSquatting, a new AI security threat that exploits how large language models hallucinate fake resource identifiers. The attack targets popular AI coding assistants including GitHub Copilot, Cursor, and Gemini CLI, with hallucination rates reaching 85% for repositories and 100% for trending skills. By registering these AI-generated fake domains, attackers can assemble massive botnets and execute large-scale attacks without targeting individual victims.
AI security paradox emerges as organizations deploy autonomous agents faster than they can govern
Organizations are racing to deploy AI agents across enterprise systems, but a troubling gap has emerged between confidence and capability. While 87% believe their identity management is ready for AI-driven automation, 46% admit their identity governance falls short. Meanwhile, autonomous cyberattacks powered by frontier AI models can now breach systems in just 27 seconds—faster than any human security team can respond.
AI system identifies people by how they walk, raising security and privacy questions
A new AI system called SKDMap-Net can identify individuals by analyzing their walking patterns using skeletal keypoints and joint movements. The technology achieved 95.8% accuracy on major datasets and works even when faces are obscured or cameras are far away. While it promises better long-range security checks for law enforcement, the behavioral biometrics approach raises significant privacy concerns about tracking public movement.
Sovereign AI
THE OUTPOST INSIGHTBrad Smith wants transparent AI rules while Krishnan rejects formal regulation entirely, exposing a split between established tech giants seeking clarity and disruptors wanting speed.
Brad Smith warns US AI policy creates uncertainty with regulation but no transparent rules
Microsoft President Brad Smith criticized Washington's approach to AI regulation, saying the US regulates AI without transparent or complete rules. Following abrupt export controls on Anthropic Fable 5 and delays to OpenAI GPT-5.6, Smith argues businesses cannot plan effectively. The moves have sparked a global push for sovereign AI as countries question reliance on American tech firms.
Prime Intellect raises $130M Series A to help enterprises build their own AI agents at $1B valuation
Prime Intellect has secured $130 million in Series A funding at a $1 billion valuation to provide enterprises with tools to train AI agents. Led by Radical Ventures with backing from Nvidia Ventures, Intel Capital, and Dell Technologies Capital, the startup has reached a $100 million annualized revenue run rate with 6,000 customers including Ramp and Zapier.
Cerebras targets Europe with multibillion-dollar AI expansion, challenging Nvidia dominance
American AI chip startup Cerebras is accelerating its European expansion with plans to deploy 200MW of AI compute capacity by 2027. The multibillion-dollar investment addresses surging demand for local AI infrastructure as European businesses and governments seek alternatives to US-concentrated computing power, while the company's stock surged 5.3% on the announcement.
Nokia Defense and NestAI advance AI defense capabilities with integrated battlefield systems
Nokia Defense and NestAI are developing their first operational AI defense capabilities following a €100 million investment announced in November 2025. The partnership creates three integrated systems combining deployable 5G networks with adaptive battlefield technology, designed to meet NATO operational requirements for European defense forces facing electronic attacks and drone threats.
AI Regulation
THE OUTPOST INSIGHTThe White House cleared GPT-5.6 for release while OpenAI offers the government a $42.6 billion stake, blurring the line between regulator and shareholder.
OpenAI launches GPT-5.6 after government testing sets new precedent for AI model releases
OpenAI will publicly release GPT-5.6 on July 9, including three variants—Sol, Terra, and Luna—after weeks of government review. The release follows a restricted preview to about 20 government-approved partners, marking the first time a US AI company has released a frontier model on the government's schedule rather than its own.
Meta tests 'super sensing' AI glasses that record continuously, plans to disable warning LED
Meta is testing prototype smart glasses that capture photos every few seconds while continuously recording audio, allowing AI to help users recall their day. The $1.5 trillion company plans to disable the recording LED indicator when these features are active, intensifying privacy concerns and triggering internal debates about wiretapping laws and biometric data regulations.
Brad Smith warns US AI policy creates uncertainty with regulation but no transparent rules
Microsoft President Brad Smith criticized Washington's approach to AI regulation, saying the US regulates AI without transparent or complete rules. Following abrupt export controls on Anthropic Fable 5 and delays to OpenAI GPT-5.6, Smith argues businesses cannot plan effectively. The moves have sparked a global push for sovereign AI as countries question reliance on American tech firms.
Preity Zinta secures court order against AI deepfakes violating personality rights
The Bombay High Court ordered Google, X, and Meta to remove AI deepfakes and morphed content featuring Preity Zinta within 72 hours. Justice Madhav Jamdar ruled that approximately 275 websites violated her personality rights, publicity rights, and moral rights under the Copyright Act, 1957. The court emphasized that these rights stem from constitutional guarantees under Article 19 and Article 21.
Edge AI
THE OUTPOST INSIGHTSunrun is turning homeowners into landlords for AI infrastructure while manufacturers see electricity bills jump 90% from the data centers it aims to replace.
Sunrun pilots AI computing in home battery systems, paying homeowners to power distributed network
Sunrun has launched a pilot program that installs AI compute nodes in homes with solar panels and battery storage, compensating homeowners for hosting the hardware. The residential solar company aims to leverage its network of over 1.1 million customers to process AI inference workloads, offering a faster alternative to traditional data centers while addressing AI's growing electricity demands.
Samsung develops Gaia chip to accelerate AI for PCs, challenging Qualcomm and Nvidia
Samsung's System LSI division is developing Gaia, a dedicated AI accelerator chip for PCs using 4nm process technology. The company has already supplied early samples to HP and Lenovo for performance testing, with mass production potentially starting in 2027. This marks Samsung's return to the PC chip market after 14 years.
Super Micro Computer launches Kubernetes Edge AI appliances with Red Hat and Everpure partners
Super Micro Computer announced new Kubernetes Edge AI appliances developed with Red Hat and Everpure, sending SMCI stock up 3.56%. The turnkey solution combines Red Hat OpenShift with Supermicro's edge computing infrastructure and Portworx data management to deploy, manage, and scale AI applications across distributed edge environments. The launch comes as short interest in the company rose to 19.42% of available float.
iOS 27 beta 3 unlocks Siri Expressive Voices and AI features across iPhone, Apple Watch
Apple released iOS 27 beta 3 with major AI upgrades including customizable Siri voices for iPhone 17 Pro models, Live Recognition accessibility tools, and Apple Intelligence integration in Home. The update also brings Siri AI to watchOS 27 and introduces new controls for AirPods Adaptive mode.
Videos
Natural Conversations with GPT-Live
Listening & Speaking with GPT-Live
What’s at the center of Claude’s mind?
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?
Synthetic Data
Synthetic data is artificially generated information created by AI rather than collected from the real world. It's increasingly used to train models when real data is scarce, expensive, or privacy-sensitive.
Top Stories
OpenAI launches GPT-Live voice models that listen and speak simultaneously for natural conversations
OpenAI unveiled GPT-Live-1 and GPT-Live-1 mini, new conversational AI models designed to make interactions feel more human. The full-duplex models can listen and speak at the same time, allowing natural interruptions and enabling features like live translation. With over 150 million people already using ChatGPT's voice features, the company is replacing its Advanced Voice Mode with these new models that can handle complex agentic tasks while maintaining conversational flow.
Videos
Natural Conversations with GPT-Live
OpenAIListening & Speaking with GPT-Live
OpenAIWhat’s at the center of Claude’s mind?
AnthropicFrom Your Feed / Also in the News
OpenAI launches GPT-Live voice models that listen and speak simultaneously for natural conversations
OpenAI unveiled GPT-Live-1 and GPT-Live-1 mini, new conversational AI models designed to make interactions feel more human. The full-duplex models can listen and speak at the same time, allowing natural interruptions and enabling features like live translation. With over 150 million people already using ChatGPT's voice features, the company is replacing its Advanced Voice Mode with these new models that can handle complex agentic tasks while maintaining conversational flow.
THE OUTPOST INSIGHTOpenAI pushes voice as the new interface while Microsoft quietly replaces its models in Office to cut costs, exposing tension in their partnership.
OpenAI launches GPT-5.6 after government testing sets new precedent for AI model releases
AI token economy splits as spending surges 10x while companies question tokenmaxxing returns
The AI token economy is experiencing a dramatic split, with commodity inference heading toward zero while frontier AI models command premium prices. Despite token prices dropping 55x since 2022, AI spending has surged 10x in recent months as companies grapple with tokenmaxxing trends that prioritize usage over output. Major tech firms are now questioning whether excessive AI token usage translates to actual productivity gains.
THE OUTPOST INSIGHTEnterprises are realizing that building the AI system around the model matters more than the model itself, just as Karp attacks token pricing and NVIDIA cuts costs 5x.
Meta invests $9.1 billion in Alberta Canada for its first AI data center, largest outside the US
Did you know?
Synthetic Data
Synthetic data is artificially generated information created by AI rather than collected from the real world. It's increasingly used to train models when real data is scarce, expensive, or privacy-sensitive.
Prime Intellect raises $130M Series A to help enterprises build their own AI agents at $1B valuation
Prime Intellect has secured $130 million in Series A funding at a $1 billion valuation to provide enterprises with tools to train AI agents. Led by Radical Ventures with backing from Nvidia Ventures, Intel Capital, and Dell Technologies Capital, the startup has reached a $100 million annualized revenue run rate with 6,000 customers including Ramp and Zapier.
THE OUTPOST INSIGHTPrime Intellect hits unicorn status selling enterprises the tools to build their own AI agents as Karp says CEOs are rejecting frontier lab pricing.
SambaNova raises $1B at $11B valuation, quintuples worth in months as AI chip race intensifies
GhostApproval flaw exposes AI coding agents to ancient Unix trick that grants attackers SSH access
Security researchers at Wiz discovered GhostApproval, a critical security vulnerability affecting six widely deployed AI coding assistants. The flaw exploits symbolic link vulnerabilities to trick agents into writing files outside their sandboxed workspace, potentially granting attackers remote code execution on developers' machines through compromised SSH keys.
THE OUTPOST INSIGHTAI coding assistants face a wave of exploits that weaponize the tools themselves—from hallucinated packages to symbolic link tricks—not the code they write.
HalluSquatting attack exploits AI hallucinations to build massive botnets through coding assistants
Meta tests 'super sensing' AI glasses that record continuously, plans to disable warning LED
Meta is testing prototype smart glasses that capture photos every few seconds while continuously recording audio, allowing AI to help users recall their day. The $1.5 trillion company plans to disable the recording LED indicator when these features are active, intensifying privacy concerns and triggering internal debates about wiretapping laws and biometric data regulations.
THE OUTPOST INSIGHTMeta just spent months forcing LED indicators on users to prevent covert recording, then built prototypes that skip the LED entirely.
Preity Zinta secures court order against AI deepfakes violating personality rights
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.
