🛡️ OpenAI unveils GPT-5.4-Cyber a week after rival's announcement of AI model ↗
OpenAI has released a cybersecurity-tuned version of its flagship model, called GPT-5.4-Cyber. It is aimed at defensive security work, which feels deliberate - and, in some respects, like a direct response to the noise around Anthropic’s Mythos.
The rollout is narrow for now. OpenAI says only vetted security vendors, organizations, and researchers will get access first, while its Trusted Access for Cyber program expands with new verification tiers that unlock stronger capabilities for vulnerability research and analysis. (Reuters)
⚙️ Meta extends custom chips deal with Broadcom to power AI ambitions ↗
Meta is doubling down on custom AI silicon with Broadcom, extending the partnership through 2029. The deal includes an initial commitment of more than one gigawatt of computing capacity, which is massive - one of those figures that sounds abstract until you pause over the sheer amount of infrastructure it implies.
The point is pretty simple: build more of the stack in-house and rely less on Nvidia’s expensive chips. Broadcom will also supply networking tech for Meta’s AI clusters, while Meta says newer MTIA chips are being lined up for inference workloads across its products. (Reuters)
🧠 Anthropic co-founder confirms the company briefed the Trump administration on Mythos ↗
Anthropic confirmed it briefed the Trump administration on Mythos, its powerful unreleased model. Jack Clark said the government needs visibility into systems like this, even as Anthropic is still fighting with the Defense Department over contracting and access - an uneasy combination, though that is the AI policy mood right now.
The company is still holding Mythos back from public release because of its claimed cybersecurity power. So this story is not chiefly about a product launch - not exactly - it is about who gets early access, who gets warned, and how close frontier AI is moving toward national security territory. (TechCrunch)
🌐 Google adds AI Skills to Chrome to help you save favorite workflows ↗
Google is adding a feature called Skills to Chrome, letting people save and reuse Gemini prompts across websites. That sounds small, maybe even a little nerdy, but it nudges the browser further toward becoming an AI workspace instead of just a tab container.
Users can save prompts from chat history, trigger them with a slash or the plus button, and run them on the page they are viewing - plus selected tabs too. Google says early uses include shopping comparisons, recipe tweaks, document summaries, and health or budgeting tasks, which feels practical with a faintly uncanny edge. (TechCrunch)
🏗️ AI data center startup Fluidstack in talks for $1B round at $18B valuation months after hitting $7.5B, says report ↗
Fluidstack is reportedly in talks to raise $1 billion at an $18 billion valuation, which is a huge leap from where it was valued not long ago. The market is still treating AI infrastructure like oxygen and gold at the same time - scarce, necessary, and extravagantly expensive.
A big reason is demand from frontier labs. Fluidstack already landed a giant deal with Anthropic to build AI-focused data centers, and that has turned it from a relatively low-profile infrastructure company into one of the hotter names in the compute scramble. (TechCrunch)
💼 Exclusive: Inside Google's AI jobs push ↗
Google is trying to shape the jobs side of the AI debate before lawmakers and labor groups define it for them. It is backing new research and training programs focused on how workers adapt, which sounds generous - and strategic too.
The programs include healthcare, manufacturing, and apprenticeship efforts, with one initiative aiming to train 40,000 workers in AI skills. Google is also convening people from government, industry, and civil society to talk through the future of work, so this is part workforce policy, part reputation management... likely both. (Axios)
FAQ
What is GPT-5.4-Cyber and who can use it first?
GPT-5.4-Cyber is a cybersecurity-tuned version of OpenAI’s flagship model, built for defensive security work. Early access is limited to vetted security vendors, organizations, and researchers rather than the general public. The rollout appears intended to keep higher-risk capabilities under tighter control while still supporting vulnerability research and analysis.
Why is OpenAI restricting GPT-5.4-Cyber access instead of launching it broadly?
The article suggests OpenAI is taking a narrow release path because cybersecurity models can be especially sensitive. By expanding its Trusted Access for Cyber program with verification tiers, OpenAI can provide stronger tools to approved users while maintaining closer oversight. That makes the launch feel more like a controlled security deployment than a standard product release.
Why does Anthropic’s Mythos matter even though it is not publicly available?
Mythos matters because it is being framed as powerful enough to trigger government briefings before any broad release. That shifts the conversation away from product marketing and toward national security, access control, and policy visibility. In practical terms, it shows how frontier AI is increasingly being treated as strategic infrastructure rather than merely commercial software.
What does Meta’s Broadcom chip deal say about the AI infrastructure race?
Meta’s extended deal with Broadcom shows how seriously major companies are investing in controlling more of the AI stack. The goal is not only to add vast computing capacity, but also to reduce reliance on Nvidia’s costly chips. It also signals that networking and inference hardware are becoming just as important as raw model development.
How big is one gigawatt of AI computing capacity in practical terms?
The article presents one gigawatt as an enormous infrastructure commitment, even if the number sounds abstract at first. In context, it points to data center scale, long-term capital spending, and a serious push to support AI workloads at massive volume. It is the kind of figure that suggests industrial-scale ambition rather than experimental deployment.
How will Google’s new Chrome Skills feature actually help everyday workflows?
Chrome Skills appears designed to make saved Gemini prompts reusable across websites and even across selected tabs. That could make routine tasks faster, such as comparing products, adjusting recipes, summarizing documents, or handling budgeting work. The larger shift is that Chrome starts acting more like an AI workspace for repeatable workflows, not just a browser.
Why is Fluidstack suddenly valued so much higher in the AI market?
Fluidstack’s reported valuation jump seems tied to how aggressively the market is rewarding companies that can supply AI infrastructure. Demand from frontier labs is a major factor, especially after its large data center deal with Anthropic. In many pipelines, access to compute has become one of the most valuable bottlenecks, which helps explain the premium.
What is Google trying to achieve with its AI jobs push?
Google appears to be moving early on the labor side of AI so it can help shape the debate before lawmakers and labor groups define it without them. The programs around healthcare, manufacturing, apprenticeships, and AI training suggest both workforce preparation and public positioning. It reads as a blend of skills investment, policy engagement, and reputation strategy.