Welcome to this edition of Loop!

To kick off your week, I’ve rounded-up the most important technology and AI updates that you should know about.

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

Let's jump in!

58% of buyers now start their research in ChatGPT or Gemini, not Google. Most startups aren't showing up there yet.

The ones that are get cited by the AI tools their buyers, investors, and future hires already use. And they convert at 3×.

Download the free AEO Playbook for Startups from HubSpot and get the exact steps to start showing up. Five minutes to read.

Get the free playbook

1. Anthropic's most powerful model won't be available to the public

We start this week with Anthropic, which has sent shockwaves through the cybersecurity world with a significantly more advanced AI model. The model is so powerful that it has already found vulnerabilities in every major operating system and web browser.

This new model is called Claude Mythos Preview and it's able to autonomously scan software systems, detect vulnerabilities, and it can even develop new exploits - without any humans being involved.

Unlike previous models, Anthropic isn't releasing this one to the public - it's only being shared with around 40 partners in the tech and security industry, including Nvidia, Apple, Microsoft, and CrowdStrike.

They hope that this will give defenders a head start by letting them find and patch weaknesses before attackers can exploit them.

If this technology is as good as Anthropic claims, it could create a huge shift in how cybersecurity works. Right now, the entire industry relies on humans and the economics have always favoured the attackers, as they only need to find one way in - whereas the defenders have to protect everything.

An AI model that can autonomously find and fix vulnerabilities changes that dynamic completely - giving defenders the kind of coverage that used to require entire teams.

Of course, this is a double-edged sword. Anthropic is restricting access for now, but this kind of technology isn't going to stay behind closed doors forever.

Other companies - including Chinese labs - will develop similar models, and once attackers have these tools too, the cyber arms race shifts to a completely different level. This means that the advantage will go to whoever has the better model, not the bigger team.

2. Intel agrees to build Elon Musk's AI chip factory

The company has signed a deal with Elon Musk to design and build his new Terafab, which will become a huge AI chip factory in Texas.

The chips are needed for both SpaceX and Tesla, as Musk pushes ahead with his wider ambitions for self-driving cars, humanoid robots, and data centres in space.

Building a chip factory requires billions of dollars, years of construction, and specialist equipment that very few companies in the world can operate - and Musk has admitted that he can't do it himself, which is why he needs Intel.

But Intel isn't in a great position either. The company has been struggling for years to compete with TSMC and is burning through a $20 billion investment on new fabs in Arizona.

So you have two parties here who badly need each other - Musk can't build the chips on his own, and Intel desperately needs big customers to make that investment worthwhile.

The big question is whether Intel can actually deliver on time, given its recent track record with delays and projects going over their budget.

3. Waymo’s robotaxis are now detecting potholes across the US

Since Waymo's robotaxis drive millions of miles across US cities every year, they've built up a detailed picture of where potholes are, how severe they are, and which ones need to be fixed urgently.

Their vehicles are already using cameras, radar, and accelerometers to detect potholes and avoid them, but Waymo is now feeding that data into Google's Waze platform - which allows city officials to access it for free.

It's a smart bit of diplomacy from Waymo, which has been facing real pushback in cities where unions are strongly opposed to robotaxis. Sharing this data costs them very little, but it gives city officials a reason to see the company as a partner rather than a threat.

Most cities still rely on residents phoning in to report potholes, so replacing that with vehicles that can detect them automatically and rank them by severity is a no-brainer.

But Waymo isn't the only company doing this. Mercedes already offers similar data from its modern cars through an API that governments can use. They've partnered with the city of London and offer this data to the council in real-time.

This is something I've actually worked on professionally. I've spoken with Mercedes about the data they collect and how this programme could be expanded for the UK Government. It's great to see Waymo do this as well, and I think this space will get a lot bigger over the next few years.

4. Google launches a competitor to Wispr Flow

Google has released a new AI dictation app called Eloquent on iOS, which allows people to interact with AI tools much faster than typing. Instead of having to type constantly, you just speak naturally and the app will transcribe everything for you - automatically removing any filler words like "um" and "ah".

It can run entirely on your phone if you're conscious about privacy, or you can switch to cloud mode for access to a more powerful model. It's free and available on iOS, with plans for Android and desktop versions, and it joins a growing list of tools in this space like Wispr Flow.

But there's a broader trend happening here. A lot of people are now becoming frustrated with AI tools and having to type constantly. That often leads to people writing shorter, more vague prompts - which means that the AI doesn't have enough context to give you a useful answer.

This cycle of typing something quickly, getting the wrong answer, and having to try again can actually hurt your productivity more than it helps.

In my own experience, voice models change that completely. I've seen a real improvement in my own work, as tools like Wispr Flow have allowed me to get things done in a few hours that would have taken double that by typing.

You can give the AI model significantly more context when you're speaking naturally, and the results are noticeably better because of it. (While Wispr Flow have sponsored me in a previous post, this is my own view and I pay to use their tool every month).

Of course, it won't work for every environment - you're probably not going to start talking out-loud in an open-plan office. But this is an incredibly powerful way for people to interact with AI tools, especially if you use them for long periods of time, and the technology has made significant progress in the last year.

5. Prediction markets are taking over US news networks

Kalshi, the US prediction market, is now going to appear across Fox's news channels after signing a new deal with the network.

If you're not familiar with Kalshi, it's a platform where you can place real money and predict what will happen with world events - including elections, unemployment figures, Oscar winners, even the Iran war.

They've already signed similar deals with CNN and CNBC, so this is now appearing across almost every major US news network. Prediction markets have made billions in revenue and they're now aggressively pushing into mainstream media, so that they can reach ordinary viewers.

Their goal is pretty simple - they want to make prediction markets feel like a normal part of watching the news, then convert as many viewers as possible into customers. Media companies are now leaning into it, as they’re seeing their ad revenues fall off a cliff and this is an incredibly lucrative deal, but the broader trend is a bit worrying.

If viewers are staking hundreds of millions of dollars on whether a news event happens or not, that creates a really uncomfortable relationship between the news and the prediction market.

Could it start to influence the editorial decisions that networks make? Trust in journalism is already at a low point, and blurring the line between news coverage and a prediction platform isn't going to help.

Microsoft releases new security tools for AI agents

Microsoft has open-sourced a new set of security tools, which are designed to monitor and control what AI agents can do while they're running.

AI agents are now doing everything from booking flights to executing trades and managing infrastructure on their own - but there's very little in place to control what they're actually doing.

Rather than building something from scratch, Microsoft has borrowed from what already works in operating systems - things like keeping processes separated, shutting them down if they go wrong, and controlling what they're allowed to access - and applied them to AI agents.

It's built to work with the frameworks developers already use, including LangChain, CrewAI, and Google ADK, so adding governance doesn't mean that you have to rewrite your agent code.

It follows OWASP's list of security risks for AI agents, which was published for the first time back in December 2025. The EU AI Act's high-risk obligations also kick in this August, which means that companies deploying agents without any governance in place are going to be in a difficult position very quickly.

I have to say, this feels like one of the more practical contributions to AI safety we've seen recently - not just guidelines or a white paper, but actual tooling that developers can quickly install and use today.

Whether teams will actually adopt it is another question. Security tooling that's optional tends to go ignored, and Microsoft knows this.

That's why they've built governance directly into the agent's execution path, so every action gets checked before it runs - rather than making it something you add on afterwards.

If you want to learn more or use this framework within your own agentic AI products, you can check out the link below.

🤖 Google's Gemini can now answer questions with 3D models and simulations

💾 The AI RAM shortage is driving up SSD prices

🍎 Apple's cheapest Mac is shaking up the PC market

🚕 VW will deploy “thousands of robotaxis” on Uber's platform in the US

🔬 Jeff Bezos's new lab hires xAI co-founder from OpenAI

📈 Samsung shares rise, after profit jumps 8-fold on AI chip boom

🚀 NASA's Artemis II crew sets a new distance record from Earth

🖱️ Sierra's Bret Taylor says the era of clicking buttons is over

🎨 Canva doubles down on AI agents with two new acquisitions

📺 YouTube eyes more interactive video formats as it grows on TV

🏗️ Firmus, the AI data centre builder backed by Nvidia, hits $5.5 billion valuation

☁️ Anthropic increases its compute spending with Google and Broadcom, amid skyrocketing demand

This Spanish startup is building satellites that will give AI companies a far more detailed view of the Earth than anything currently available. This data is incredibly valuable for training new AI models, as it allows them to be far more accurate at things like predicting weather patterns, monitoring crops, and tracking changes to infrastructure.

Xoople (pronounced "zoople") was founded in 2019 and has used data from government satellites to build its own technology. The startup has also integrated with cloud platforms like Microsoft and Esri, which is where most of its potential customers already work.

To get the satellites off the ground, Xoople has just closed a $130 million Series B, bringing its total funding to $225 million. They've also signed a deal with US defence contractor L3Harris to build the sensors for its satellites.

What makes Xoople's approach clever is that most Earth observation companies launch satellites first and figure out distribution later. Xoople has done the complete opposite, and it's already built into the platforms that enterprise and government buyers use every day.

Until its own satellites are ready, it's making good use of publicly available imagery from sources like ESA's Sentinel-2.

It's a smart strategy, and I think there's a lot of potential here. That being said, they're entering a very crowded market where competitors like Planet and Vantor already have satellites that are collecting vast amounts of data, so the pressure is on Xoople to prove that its data will be worth the wait.

If you want to find out more about the company, I've included a link below.

This Week’s Art

We’ve covered quite a bit this week, including:

If you found something interesting in this week’s edition, please feel free to share this newsletter with your colleagues.

Or if you’re interested in chatting with me about the above, simply reply to this email and I’ll get back to you.

Have a good week!

Liam

Share with Others

If you found something interesting in this week’s edition, feel free to share this newsletter with your colleagues.

Liam McCormick is a Senior AI Engineer and works within Kainos' Innovation team. He identifies business value in emerging technologies, implements them, and then shares these insights with others.