Crypto impersonation scams grew 1,400% as AI supercharged fraud
Impersonation scams increased 1,400% year over year in 2025, with average payments rising over 600%. The average crypto payment across all scam categories climbed from $782 to $2,764.
SpaceXAI's Grok Bot turns agents into persistent digital coworkers that can operate your apps for $120-per-month
Persistent AI agents called Grok Bots can operate software applications continuously and autonomously across multiple platforms for $120 per month. Each Bot accesses applications through its own computer environment, continues working when the user's laptop is closed, and returns when needing approval or completing assignments.
SpaceXAI launches Grok Bot as the agent race moves to office work
SpaceXAI released Grok Bot, AI agents that sign into apps and websites while retaining context between tasks and coordinating with each other. Initial access is limited to SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium subscribers, with an enterprise waitlist available.
Anthropic models will soon inject watermarks identifying AI-generated text
Anthropic will embed invisible watermarks in all text generated by Claude models starting August 2, 2026, to identify AI-generated content across all countries. The watermark persists when text is copied and pasted, and earlier models must be upgraded by December 2 to comply with EU AI Act transparency requirements.
IBM bets $240m on cheap, open-source inference to take on the hyperscalers
IBM is investing $240 million in a multi-year deal with startup Together AI to deploy Nvidia Blackwell systems on IBM Cloud, betting that enterprises prioritize inference cost over model prestige. The partnership focuses on making cheaper open-source AI inference available to enterprises through IBM Cloud infrastructure.
Mistral AI wants to build 1 gigawatt of European compute by 2030 — and lock in customers now.
Mistral AI announced plans to build one gigawatt of European compute capacity by 2030 through multi-year customer commitments, with 200 megawatts operational by 2027. The company is also introducing regional inference endpoints, uptime guarantees for mission-critical workloads, and hosting third-party open models like GLM-5.2 from Chinese lab Z.ai.
LTX-2.5 can generate a 10-second AI video from an image in just 6.8 seconds on Nvidia superchips — and it's open weights
LTX-2.5 generates a 10-second AI video from a single image in 6.8 seconds on Nvidia superchips. The open-weights model is now available on Hugging Face, integrated into ComfyUI, and accessible through the LTX API for teams.
Nvidia's Switchyard router reshuffles AI models mid-task, cutting task costs to a third in its own tests
Nvidia released Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model, alongside NeMo Switchyard, a routing library that directs each workflow step to the most appropriate model. According to Nvidia's tests, the combination achieves frontier-level task completion while reducing costs to approximately one-third compared to running Opus 4.8 alone.
Corma raised $60M from Sequoia to build the defensive AI that cybersecurity is missing
AI models successfully planted persistent threats in 88% of simulations testing Fortune 500 company defenses. Corma raised $60 million from Sequoia to develop defensive AI that identifies security vulnerabilities by simulating attacks across dozens of security tools.
Point2 raised $136M to solve AI’s wiring problem. Its RF cables beat copper on reach and optics on latency.
RF cables that outperform copper on range and optics on latency are being commercialized after Point2 Technology raised $136 million in Series B funding. The cables are designed to replace both copper and traditional optical connections inside AI data centres.
OpenAI is hiring a power trader, because electricity is now a position on its books
OpenAI is hiring a power trader to manage its electricity trading and commodity hedging strategy across data centers. The new role will initially have no direct reports and represents the company treating electricity as a significant business position.
Mark Zuckerberg lays out Meta’s AI vision in a 6,500-word essay: 6 things to know
Meta announced a $1 billion fund to invest in communities where its data centers operate and introduced its new open-weight AI model called Muse Glimmer. Zuckerberg argued in his 6,500-word essay that open-source AI models prevent harmful centralization and empower users more effectively than closed systems.
Nvidia Taps Wall Street for $500 Billion Funding Commitment
Nvidia secured a $500 billion financing commitment from Apollo Global Management, Blackstone, BlackRock, and Brookfield Asset Management for artificial intelligence infrastructure. The funding will support the development and deployment of AI computing systems.
Sony, TSMC to Invest $6.4 Billion in Japanese Sensor Plant
Sony and TSMC plan to invest $6.4 billion in a Japanese image sensor factory. The combined investment of ¥1 trillion will fund their planned manufacturing facility in Japan.
Google’s top hacker hunter explains why hacking groups get codenames
Google changed its naming convention for hacking groups instead of using traditional animal-based or mythological names. The company now uses more descriptive names based on the groups' characteristics, locations, or tactics they employ.
Singapore to Ensure AI Helps Workers Worried About Jobs, PM Says
Singapore's government is using artificial intelligence to increase worker productivity and create better jobs as AI rapidly expands. Prime Minister Lawrence Wong stated this approach addresses worker concerns about their livelihoods amid technological change.
No cloud, no GPUs, no problem: Liquid AI's new model LFM2.5-2.6B brings powerful AI agents to devices as small as a Raspberry Pi
Liquid AI released LFM2.5-2.6B, a 2.6 billion parameter language model designed to run on local devices including Raspberry Pi without cloud or GPUs. The model handles agentic tasks like tool calling, document management, and workflow automation in connectivity-limited environments.
China’s Unitree Seeks $904 Million in First Mainland Robotic IPO
Unitree Robotics seeks to raise 6.1 billion yuan through an initial public offering that would establish it as mainland China's first publicly traded humanoid robot maker. The IPO valuation would represent a significant milestone for the robotics industry in China's capital markets.
Why Lightspeed is going all-in on creator-led venture capital
Lightspeed Venture Partners hired Claire Zau, a seed investor with a major Instagram following, to build trust with next-generation founders. The move follows similar investments by a16z in Erik Torenberg's Turpentine podcast and OpenAI in TBPN, reflecting a trend of venture firms leveraging creators' influence.
Anthropic confirmed it is designing custom chips for Claude. It wants engineers who have “shipped silicon.”
Anthropic is designing custom silicon chips for its Claude AI model to improve performance and efficiency at scale. The company is building an internal team and seeking engineers with experience shipping silicon products.
Yann LeCun’s second attempt at a VC fund lands with $100M and Gemini’s co-lead as a partner
A venture capital fund called 224 Ventures launched with over $100 million in assets under management, with Yann LeCun and Oriol Vinyals as partners. Vinyals previously served as Gemini's co-technical lead at Google DeepMind before joining the new firm.
A startup that slashes AI agent bills just raised $35M. Anthropic is a backer.
Sapiom, a startup positioned between AI agents and the models they operate on, raised $35 million in Series A funding led by Dragonfly. The round brings total funding to $50 million across a seed and Series A completed within 17 months of the company's launch.
Google is cutting 52 jobs in Washington the same week it lost four star researchers
Google is permanently laying off 52 employees in Washington state. The company provided no reason for the cuts and stated they did not result from work relocation or outsourcing.
Anthropic built an inspection layer that lets enterprises block sensitive data before it reaches Claude
Anthropic launched inference hooks for Claude Enterprise that route every prompt through a customer's security server for inspection before Claude processes it. The system checks both prompts and tool calls, blocking sensitive data if the security server returns a deny verdict.
Reddit’s new Rules Hub uses AI to enforce moderation by intent, not keywords. Automod’s other features stay.
A moderation system called Rules Hub uses large language models to evaluate whether posts match community rule intent rather than keywords. The system works alongside Post & Comment Guidance and Safety Filters to replace Automod's enforcement capabilities while keeping Automod's other features.
Google named the new boss of Gemini 4. Both of Gemini’s co-leads quit the same day
Jeff Dean and Sanjay Ghemawat departed Google after 27 years to start a new company called Discovery Loop. Google appointed a new leader for Gemini while two co-leads of Gemini also quit the same day.
Google is killing Assistant on phones September 4. Gemini is the only option left.
Google is discontinuing Assistant on Android phones starting September 4, replacing it entirely with Gemini. The shutdown process will take several weeks to complete across all devices.