🛡️ Codex Security: now in research preview ↗
OpenAI rolled out Codex Security in research preview for ChatGPT Pro, Enterprise, Business, and Edu users through the Codex web app. The pitch is fairly direct - let an AI agent find and fix app security issues before they turn into a serious headache. (OpenAI)
It is launching with free usage for the first month, which feels like a deliberate push to get teams experimenting quickly. Security copilots are starting to look like a small arms race of their own now. (OpenAI)
🇬🇧 UK should back licensing-first approach for AI training, says upper house committee ↗
A UK House of Lords committee said the government should prefer a licensing-first model for AI training, rather than leaning on broader exceptions that could let companies use copyrighted material more freely. That is a major signal in the copyright fight - perhaps not flashy, but highly consequential. (Reuters)
The recommendation adds pressure on ministers to show they are not simply clearing the runway for AI firms at creators’ expense. In a quiet way, this may shape AI behavior more than half the splashy model launches do. (Reuters)
🏗️ Oracle and OpenAI drop Texas data center expansion plan, Bloomberg News reports ↗
Oracle and OpenAI have reportedly scrapped plans to expand a Texas data center site. That does not mean the infrastructure boom is over - far from it - but it does suggest that even the largest AI buildouts are running into hard constraints, recalculations, or both. (Reuters)
With AI demand still surging, any retreat from a major compute project stands out. It is a reminder that the future of AI is not just models and demos - it is land, power, cooling, contracts... the supposedly dull stuff that suddenly feels anything but dull. (Reuters)
📈 Marvell Tech shares soar on bullish multi-year AI chip outlook ↗
Marvell jumped after giving an upbeat multi-year forecast tied to demand for custom AI chips from big tech customers. Investors clearly heard what they wanted to hear: AI infrastructure spending is still very much alive, perhaps a bit uncomfortably so. (Reuters)
The move also reinforces how much of the AI story sits below the chatbot layer. Chips, networking, and custom silicon keep hoovering up attention - and money - because that is where much of the leverage sits. (Reuters)
🇨🇳 China's largest provincial economy vows to reshape industry with AI ↗
Jiangsu, China’s largest provincial economy, said it wants to use AI to reshape industry. The emphasis is not just on flashy consumer products but on industrial deployment, which is less glamorous, certainly, yet probably where much of the economic shift will land. (Reuters)
The announcement fits a broader Chinese push to move AI deeper into manufacturing and core sectors. So yes, another policy story - but one with serious weight, because scale in Jiangsu is like steering a cargo ship with a neural net strapped to it. A peculiar image, perhaps, though it captures the point. (Reuters)
💬 After Europe, WhatsApp will let rival AI companies offer chatbots in Brazil ↗
Meta is opening WhatsApp in Brazil to rival AI chatbot providers for a fee, following a similar move in Europe. That is notable because it nudges WhatsApp from being merely a Meta AI surface into something more like a distribution layer for multiple AI vendors. (TechCrunch)
It is also a quiet sign that platform control and AI reach are becoming tightly entangled. The chatbot wars are no longer just about who has the smartest model - they are also about who owns the doorway. (TechCrunch)
🦊 Partnering with Mozilla to improve Firefox's security ↗
Anthropic said Claude Opus 4.6 worked with Mozilla researchers and found 22 Firefox vulnerabilities over two weeks, with Mozilla classifying 14 of them as high severity. That is a substantial result, and it gives the "AI for security" pitch more substance than the usual vague promises. (Anthropic)
The broader takeaway is that AI tools are starting to prove their worth in narrow, high-value workflows where speed matters and the output can be checked by humans. Less magic robot, more very caffeinated bug hunter. (Anthropic)
FAQ
What is Codex Security and why does it matter for app security teams?
Codex Security is presented as an AI agent that can find and fix application security issues before they become bigger problems. In this roundup, it launched in research preview for ChatGPT Pro, Enterprise, Business, and Edu users through the Codex web app. The first month of free usage also suggests OpenAI wants teams to test it quickly in real workflows.
Why is the free first month for Codex Security such a big deal?
A free first month lowers the barrier for security teams to experiment without immediate budget pressure. That kind of launch tactic usually helps vendors gather usage, feedback, and early case studies fast. In practical terms, it turns curiosity into trials and could accelerate adoption if teams see useful results early.
What does the UK licensing-first proposal mean for AI training and copyright?
The House of Lords committee is signaling that AI training should rely more on licensing than on broad copyright exceptions. That matters because it pushes policy toward negotiated access instead of freer use of protected material. For creators and AI companies, the recommendation could shape how training data is sourced, priced, and disputed.
Why does this AI news roundup focus so much on infrastructure, not just chatbots?
Several items point below the chatbot layer to the systems that make AI possible at scale. The Texas data center reversal highlights land, power, cooling, and contracts, while Marvell’s forecast points to chips and networking as major leverage points. Together, they show that AI growth depends as much on physical infrastructure as on model demos.
Why would OpenAI and Oracle dropping a Texas data center expansion plan matter?
A canceled expansion stands out because AI demand is still described as strong. It suggests that even very large AI infrastructure projects can hit practical constraints or require strategic recalculation. The bigger takeaway is that growth in AI is not frictionless, especially when compute buildouts depend on complex real-world resources.
What does Marvell’s surge say about where AI money is flowing?
Marvell’s rally reflects investor confidence in continued demand for custom AI chips from major technology customers. That reinforces the idea that a large share of AI spending is still concentrated in hardware, networking, and infrastructure. Even when consumer-facing AI gets the headlines, the deeper revenue story often sits in the systems underneath.
Why is China’s Jiangsu province AI push important for the broader AI market?
Jiangsu is described as China’s largest provincial economy, so an industrial AI push there carries real scale. The emphasis is on reshaping manufacturing and core sectors rather than only building flashy consumer products. That matters because much of AI’s economic impact may come from quieter deployment inside industry, logistics, and production workflows.
How does WhatsApp opening to rival chatbots in Brazil change the AI platform race?
It suggests WhatsApp is becoming more than a home for Meta AI and more like a distribution channel for multiple AI providers. That shift matters because control of user access can be as important as model quality. In many AI markets, owning the doorway to users may shape competition just as much as having the smartest assistant.