Brave is changing how its search results handle clicks to make large-scale scraping harder. Encrypted result links let Brave register that a result was opened without identifying which one; bulk collectors would need many keys, and Brave can count requests. The company says the approach should reduce server strain and CAPTCHAs for ordinary users. It requires JavaScript, and copied redirect links may expire, so they are not suitable as bookmarks.
Several companies are building AI infrastructure and tools. Mount Thor raised $45 million in seed and Series A funding to expand its dedicated Macs and macOS virtual machines for research labs, sandbox providers and agent products; it also raised undisclosed non-dilutive hardware capital. It plans an inference lab for open-source models optimized for Apple silicon, but has not released pricing or performance comparisons. Intel and RadixArk, meanwhile, are optimizing the open-source SGLang inference system for Intel CPUs, GPUs and accelerators, including Advanced Matrix Extensions, with goals of lower latency and more efficient deployment.
Collaboration app Delta, demonstrated by Richard Feldman, keeps agent conversations, code changes and reviews in shared threads across Mac, Linux, Windows and mobile. Teammates can comment on code, send feedback to an agent, and approve or reject changes. Terminal output syncs, but viewing it remotely is read-only to prevent remote code execution. Delta supports different models and agent harnesses; its free private beta has only been tested with groups of up to 30–40 people, and pricing is undecided. An on-premises pilot is underway for regulated users.
OpenAI is committing $2 million to research AI safety for children, with findings—including negative or inconclusive results—required to be published under the researchers’ control. The funding will not directly change children’s chatbot experience. OpenAI also cites default age-appropriate safeguards, parental tools and regular risk assessments. European policymakers are expected to consider whether voluntary commitments should become enforceable checks in discussions around the EU Kids Act.
Bright Machines says its system can move live production data from air-gapped factories to the cloud without permitting inbound network access. Operators use an on-premises dashboard while an edge writer copies Iceberg objects; AWS Glue registers tables only after metadata replication is complete, and Unity Catalog governs validated data. Compared with an earlier nine-service setup that could miss schema changes, a five-minute Debezium heartbeat now flags sources more than 30 minutes behind.
Railway reported that attackers exploited build-infrastructure vulnerabilities in an attack lasting nearly 26 hours, from October 20 to 21, 2026, and used stolen credentials to incur about $740,000 in AI usage charges. The attackers decrypted a subset of customer variables, affecting at most 1.85% of workspaces—roughly 100,000 credentials—and accessed linked-account tokens for more than 1,850 users. Railway says it reset tokens, worked with providers to revoke exposed keys and is pursuing customer reimbursements. It urged users to rotate keys, check audit logs and remove public database access.
Security concerns are also drawing attention to hiring by AI companies: the article says firms are recruiting former intelligence officers to guard against espionage. Reports allege two extramarital affairs involving Nvidia CEO Jensen Huang, including one with an alleged Chinese spy; Nvidia’s board hired a law firm to investigate. Nvidia has not commented. Its shares reportedly fell more than 50% after the news and traded around $200, compared with a January high above $400, while the company was discussing possible layoffs.
Forecasts for AI infrastructure spending have climbed sharply. Estimates for 2026 spending by Alphabet, Amazon, Meta, Microsoft and Oracle rose from $546 billion in January to $800 billion; the 2027 forecast increased from $630 billion to $1.1 trillion. Morgan Stanley sees spending approaching $1.6 trillion in 2027 if CoreWeave, Nebius and Space Exploration Technologies are included. The article says the spending could benefit Nvidia, Micron, Broadcom and AMD, with Micron supplying high-bandwidth memory used alongside AI processors. Separately, the UK’s Sovereign AI Venture Fund and British Business Bank backed Fractile in a Series C round worth more than £500 million, valuing the London- and Bristol-based chip startup at over £5 billion. CapitalG, Lightspeed and Redpoint also joined the round.
Enterprise adoption remains ahead of measurement, according to Genpact’s November 12, 2026 report, based on more than 2,000 executives across 17 industries. It found that 79% of enterprises have AI agents in production, but only 20% use them broadly across functions; just 15% link AI investment fully to profit-and-loss outcomes. On average, 49% of AI projects funded in the past 24 months delivered expected value. Genpact classified 17% of enterprises as AI Leaders, and found that 92% of companies with agents in production kept their workforce the same size or expanded it.
Nvidia introduced Scale In to help secure and operate networks of AI agents in AI factories. The system combines BlueField 4 and DOCA with Spectrum X Ethernet for security monitoring and data movement. Nvidia says BlueField 4 can monitor activity across memory, storage, GPUs and CPUs, while Nvidia Sentry can monitor agent behavior and quarantine agents that cross their defined software boundaries. The tools build on Nvidia’s Open Agent Safety Platform, introduced in September, and OpenShell Reference Architecture, announced in October. In markets, AI-related uncertainty followed a volatile Wall Street session: Australia’s S&P/ASX 200 was down 35.3 points, or 0.4%, to 9,117.9 at 10:03 a.m. AEDT Thursday. Five of 11 sectors opened higher, Xero rose nearly 10%, and mining stocks weighed on the broader index.
Key Takeaways
- OpenAI committed $2 million to independent, publishable research on AI safety for children; the funding does not directly change chatbot experiences.
- Mount Thor raised $45 million to expand Apple-based AI infrastructure, while Intel and RadixArk are optimizing SGLang for Intel CPUs, GPUs and accelerators.
- Nvidia’s Scale In combines BlueField 4, DOCA, Spectrum X Ethernet and Nvidia Sentry to monitor infrastructure and agent behavior.
- AI infrastructure spending estimates for five hyperscalers reached $800 billion for 2026 and $1.1 trillion for 2027; Morgan Stanley sees nearly $1.6 trillion in 2027 including additional firms.
- UK-backed Fractile raised a Series C round worth more than ÂŁ500 million at a valuation above ÂŁ5 billion to develop AI chips and related technology.
- Railway says a 2026 breach exposed up to roughly 100,000 credentials and linked-account tokens for more than 1,850 users, generating about $740,000 in AI charges.
- Genpact found 79% of surveyed enterprises have AI agents in production, but just 15% fully connect AI investments to profit-and-loss outcomes.
- Bright Machines says it can replicate data from air-gapped factories to cloud systems without inbound network access, using validation and monitoring steps.
- Delta’s shared agent-work threads support code review and team feedback, but the beta has only been tested with groups of up to 30–40 people.
- Brave is encrypting search-result links to make bulk scraping harder; Australia’s S&P/ASX 200 fell 0.4% amid AI-trade uncertainty.
Brave hides search clicks to slow down scrapers
Brave says it will hide which search result a person clicks to make large scale scraping more difficult. Each result link uses encryption so Brave can tell that someone opened a result on a page, but not which result they chose. Scrapers need many keys, and Brave can count requests to make bulk collection expensive. Brave says the change should reduce the extra server load and CAPTCHAs that hurt real users. The system needs JavaScript, and copied redirect links may expire, so they should not be used as bookmarks.
Delta lets teams share full AI agent threads
Richard Feldman showed Delta, a collaboration app for Mac, Linux, Windows and mobile that keeps chat, code changes and reviews in one shared thread. Agent transcripts stay attached, and line level comments on code can go directly to the agent; teammates can also approve or reject changes. Delta syncs terminal output but makes terminal viewing read only to prevent remote code execution. Each person gets a local checkout from a Delta work tree and can use a different agent, while the system works with different models and harnesses. Feldman said it had not been tested at large scale, beyond standups of 30 to 40 people; its private beta is free, pricing is undecided, and an on premises pilot is underway for regulated users.
OpenAI commits $2 million to research AI safety for children
OpenAI is putting $2 million into research on AI safety for children. The article says the funding will not directly change what children experience in chatbots. OpenAI says research teams will control their conclusions and must publish their findings, including negative or inconclusive results. The company also points to age appropriate safeguards by default, parent tools, regular risk assessments and accountability. European policymakers are expected to discuss whether voluntary commitments should become enforceable checks through the EU Kids Act talks.
Mount Thor raises $45 million for Apple based AI infrastructure
Mount Thor raised $45 million through Seed and Series A financings to expand its AI infrastructure using Apple hardware. It also raised undisclosed non dilutive capital for hardware purchases, but did not share how the equity funding splits between the rounds. The company serves research labs building reinforcement learning environments, macOS sandbox providers and agent products that need persistent machines. Mount Thor offers dedicated Macs and macOS virtual machines, managed through a portal, command line tool and APIs. It also plans an inference lab for open source models optimized for Apple silicon, but has not provided performance comparisons or prices.
Intel and RadixArk tune SGLang for Intel AI systems
Intel and RadixArk are working to optimize SGLang, an open source AI inference system, for Intel platforms. The effort aims to help developers and companies run leading AI models and build applications more efficiently and affordably. The companies say the work will support Intel CPUs, GPUs and accelerators, including Intel Advanced Matrix Extensions. They aim to reduce latency, speed up deployment and give developers flexible choices across existing data centers and cloud capacity. RadixArk says it will contribute optimizations to SGLang’s core framework.
Bright Machines brings live data out of air-gapped factories
Bright Machines has built a way for air-gapped factories to share live production data without allowing inbound network access. Its system keeps operators on an on-premises dashboard while an edge writer copies Iceberg objects to a cloud account. AWS Glue registers tables only after their metadata has fully replicated, and Unity Catalog governs the validated data. The earlier system used nine services and could miss schema changes, while a five-minute Debezium heartbeat now flags sources that fall more than 30 minutes behind.
Railway hackers steal AI credentials and rack up $740,000 in charges
Attackers exploited vulnerabilities in Railway’s build infrastructure, then used stolen credentials to generate about $740,000 in AI usage charges. Railway said the nearly 26-hour attack began on October 20, 2026, and the attackers were removed on October 21. They obtained and decrypted a subset of customer variables, affecting at most 1.85% of workspaces, or roughly 100,000 credentials, and accessed linked-account tokens belonging to more than 1,850 users. Railway reset tokens, worked with providers to revoke exposed keys, and said it is working to reimburse customers. It advised users to rotate keys, review audit logs, and remove public database access.
AI companies hire former spies as Nvidia allegations raise security concerns
AI companies are hiring former intelligence officers to guard against espionage, as allegations involving Nvidia CEO Jensen Huang draw attention to security risks in the industry. The article says reports describe two alleged extramarital affairs involving Huang, one with an alleged Chinese spy. Nvidia’s board hired a law firm to investigate, and the company is discussing possible layoffs. Its shares fell more than 50% after the news and were trading around $200, down from a January high above $400. Nvidia has not commented on the allegations.
AI spending forecasts rise, lifting prospects for chipmakers
Wall Street has sharply raised its estimates for AI infrastructure spending by the five largest hyperscalers: Alphabet, Amazon, Meta Platforms, Microsoft, and Oracle. Their expected 2026 spending rose from $546 billion in January to $800 billion, while the 2027 estimate increased from $630 billion to $1.1 trillion. Morgan Stanley sees spending nearing $1.6 trillion in 2027 if CoreWeave, Nebius, and Space Exploration Technologies are included. The article says this higher spending could benefit Nvidia, Micron Technology, Broadcom, and Advanced Micro Devices. Micron makes memory chips, including high-bandwidth memory used with AI processors.
UK backs Fractile in a funding round worth over ÂŁ500 million
Fractile, a London and Bristol startup developing AI chips and related technology, received backing from the UK Government’s Sovereign AI Venture Fund and the British Business Bank as part of a funding round worth over £500 million. The Series C round also includes CapitalG, Lightspeed, and Redpoint, and values Fractile at over £5 billion. The company aims to make AI hardware faster, more energy efficient, and less costly by working across chips, software, and machine learning research. The UK government says the investment will support domestic AI hardware, reduce reliance on overseas supply chains, and help Fractile grow in Britain.
Australian Shares Fall as AI Uncertainty Weighs on Markets
The S&P/ASX 200 fell 35.3 points, or 0.4 percent, to 9,117.9 by 10:03 a.m. AEDT on Thursday. The decline followed a volatile night on Wall Street and uncertainty around the artificial intelligence trade. Five of the 11 sectors opened higher, with Xero Limited shares rising nearly 10 percent. Mining stocks fell and weighed on the wider market.
79 Percent of Enterprises Use AI Agents, but Few Track Their Business Value
Genpact released its report, Autonomy in Practice, on Nov. 12, 2026, based on a survey of more than 2,000 executives across 17 industries. It found that 79 percent of enterprises have AI agents in production, but only 20 percent use them broadly and consistently across functions. Just 15 percent have fully linked AI investment to profit and loss outcomes, and 49 percent of AI projects funded in the past 24 months delivered their expected value on average. Genpact identified 17 percent of enterprises as AI Leaders, which are further along in building an operating model that brings people and AI agents together. Among companies with agents in production, 92 percent kept their workforce the same size or grew it.
Nvidia Adds Scale In Security Tools for AI Agent Networks
Nvidia introduced Scale In to help secure, manage and operate agentic AI factories as AI agents take on larger roles. The system uses Nvidia BlueField 4 and DOCA, connected through Nvidia Spectrum X Ethernet, to provide security monitoring and data movement. Nvidia says BlueField 4 can monitor activity across memory, storage, GPUs and CPUs, while Nvidia Sentry can monitor agent behavior and quarantine agents that leave their defined software boundaries. The tools build on Nvidia’s Open Agent Safety Platform, introduced in September, and the OpenShell Reference Architecture announced in October.
Sources
- Brave will hide clicks to starve scrapers
- Share the Agent Thread, Not the Transcript
- OpenAI puts $2M into kids AI safety research
- Mount Thor Raises $45 Million to Bring Apple Hardware to Agentic AI Infrastructure
- Intel, RadixArk Collaborate to Advance Open AI Inference on Intel...
- Air Gaps Don't Have to Mean Dark Data
- Hackers stole AI credentials to get $740,000 in free compute; Railway picked up the bill
- Help wanted: AI firms court ex-spies as Nvidia drama spotlights security risks
- Great News for Nvidia, Micron, Broadcom, and AMD Stock Investors -- Wall Street Made a Mistake
- Great Britain: Sovereign AI And British Business Bank Invest In British Firm Developing The Next Generation Of AI Hardware
- ASX Opens Lower Amid AI Trade Uncertainty
- 79% of Enterprises Report AI Agents in Production, but Only 15% Have Fully Tied AI Investment to P&L Outcomes
- AI networking adds security: Nvidia unveils Scale-In
Comments
Please log in to post a comment.