You’re receiving this email because you subscribed to the newsletter. If you’d like to unsubscribe, you can manage your preferences here with one click (or use the button at the bottom of the email).

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!

1. We are losing the battle against deepfakes

We start this week with the rise of deepfakes, which are becoming difficult to spot, much easier to create, and are now a threat to our collective sense of reality. In a remarkable admission, Instagram’s CEO has said that people should no longer assume photos and videos are real by default.

In recent years, tech giants have released tools that try to identify deepfakes and AI-generated images - but these efforts seem to be failing. C2PA was one of those initiatives, which I’ve covered previously.

This new standard - which was promoted by Adobe and backed by Meta, Microsoft, and OpenAI - was supposed to let platforms flag whether an image or video was real or AI-generated.

But adoption has been half-hearted. Social media platforms will often remove the metadata when a file is uploaded. Even when they don't, they struggle to interpret it properly.

There’s also a structural problem that’s limiting efforts to identify and stop deepfakes - the companies that are investing heavily in AI are the ones that own the social media platforms.

By labelling lots of content as AI, it could undercut their own R&D investments and lead to less engagement from users - directly impacting the tech company’s ad revenues.

With no universal fix on the horizon, the next move likely falls to regulators. But as things stand, the industry's response amounts to joining initiatives, issuing vague reassurances, and hoping that someone else will figure it out.

2. OpenAI and Anthropic release new advanced AI models

It seems like there’s a new AI model every week, so it can feel hard to keep up with the latest changes - I hope this newsletter makes it a little bit easier for you.

Both companies have made steps forward with their most advanced models, which are showing steady improvements and can handle more difficult tasks.

Anthropic has launched Claude Opus 4.6, which is better at producing high-quality responses at the first attempt - whether it’s for documents, spreadsheets, presentations, or coding.

It costs the same as the previous version and is available immediately, but its context window now supports 1 million tokens - which is a huge leap over the previous models. It’s available via the API and you will need to pay more when you go over 200k tokens, which is similar to Google’s approach.

What's more interesting is the strategic direction. Anthropic is clearly looking to push Claude beyond its coding-heavy reputation and into broader knowledge work.

The company has invested in making the model better at creating PowerPoint presentations and Excel documents, and used the launch to promote Cowork - its friendlier alternative to Claude Code that’s aimed at non-technical users in marketing, research, and similar fields.

But OpenAI has also released its latest model - called GPT-5.3-Codex - which is 25% faster than before and is more efficient at using tokens. You should notice this improvement when you’re running long, complex tasks.

It shows impressive scores on industry benchmarks, similar to Anthropic’s Opus 4.6 model. Overall, you should view these are nice performance improvements for difficult tasks and may allow some new use cases to open up. It’ll take time before we can digest these changes and spot new use cases, but these are welcome upgrades.

3. Intel will make its own GPUs and compete with Nvidia

The struggling company has announced plans to compete with Nvidia and develop its own GPUs, which are the specialised chips that allow us to train new AI models and run high-performance games.

It's a significant departure for a company that has built its success on CPUs, and one that raises eyebrows given Tan's earlier promise to consolidate and focus on core businesses when he took over last March.

The initiative appears to be in its early days, as Tan indicated that their strategy will be shaped around customer demand - rather than following a fixed roadmap.

Quite frankly, it’s incredible that it took them this long to make a decision. Nvidia has seen its revenues skyrocket in the last few years, as companies invest significantly more in AI models.

To put this into perspective, Nvidia’s revenues in Q3 2021 were $7.1 billion - it then reached a whopping $57 billion just four years later in Q3 2025. I’m genuinely surprised they didn’t make this decision much sooner, as no major chipmaker can afford to sit on the sidelines.

4. OpenAI announces a single platform to control your AI agents

Frontier is a new platform from OpenAI and will allow businesses to build, deploy, and manage their AI agents in one place.

The platform will sit above your existing tools and should create a single place for agents to communicate, access shared business context, and run across different environments.

It’s great news for businesses, which quickly run into this problem - as they often have a huge number of workflows, siloed data, and legacy systems that are difficult to change.

Uber, Intuit, and HP have announced that they will become early adopters, although the platform won’t become generally available for several months.

It will act as a direct competitor to Microsoft’s Agent 365 platform and Claude’s Cowork tool, as they also allow businesses to create their own agents and store data in one place.

5. SpaceX and xAI merge to “build data centres in space”

Elon Musk announced that he's merging two of his companies, SpaceX and xAI, in a deal valued at $1.25 trillion. According to Musk, this is being done to create “AI data centres in space” - which I’ve briefly talked about before.

Google is also looking into this, alongside Nvidia, and companies both in China and Europe. It’s hoped that space data centres will become easier to manage and use less power for cooling, compared to those on Earth.

Of course, that’s the headline. What happens in reality is quite different, as Musk has claimed many things over the years - remember when he said that Tesla will have “fully autonomous” cars in 2016? Or that they’ll be able to “drive from LA to Times Square”, without you having to touch a button?

Instead, this seems to be related to SpaceX’s upcoming IPO and xAI burning money. The rocket company made roughly $8 billion last year, while xAI is spending $1 billion a month to compete with OpenAI and Google.

It’s worth noting that Musk has done this before. In 2016, he merged the debt-laden SolarCity with Tesla in what shareholders called a "bailout”.

As for space-based data centres, the engineering challenges remain enormous. Powering GPUs in orbit, managing latency, and making the economics work are all unsolved problems.

It's classic Musk futurism - bold, attention-grabbing headlines, and success is unlikely for several more years.

Waymo uses world models to train autonomous vehicles

There was an interesting announcement from Waymo over the weekend, which has used Google’s Genie 3 model to create its own world model. If you’re unsure what a world model is, it allows us to generate 3D worlds and incredibly accurate simulations - using simple text prompts or an image you uploaded.

Waymo has been using this technology to train its autonomous vehicles, allowing it to simulate thousands of different scenarios - such as a child walking in front of the car, different rain patterns, or icy conditions.

For example, engineers could ask how their autonomous vehicle would handle a snowstorm on the Golden Gate Bridge - or a hurricane in Florida.

Importantly, the model can generate both camera and lidar data at the same time - so it simulates the exact same data that’s collected in the real-world.

But the real value here is that Waymo can convert any dashcam video into a complete simulation, allowing engineers to scan the web and create significantly more training data for their AI system.

This allows them to reach significantly more scale than before and provide more scenarios for the AI to learn from, even if their Waymo vehicles haven’t encountered it before.

This breakthrough will support their global expansion plans - which include the UK, Australia, and Japan - and more quickly understand the highway code for each nation.

As I’ve said before, world models are the next big thing and will change how we train future AI systems.

🎬 Chinese company releases an incredibly realistic video generator, called Kling 3

🏢 NY lawmakers want to pause new data centres for 3 years

🇪🇺 EU orders TikTok to disable its "addictive features"

🚗 Stellantis has lost $26 billion on its electric vehicles

💰 Amazon will spend $200 billion in 2026

📊 Google earned $400 billion in annual revenue, for the first time

🔒 OpenClaw's AI skill extensions are a security nightmare

💻 GitHub and Xcode now support Claude and Codex AI coding agents

🇪🇸 Spain could block teens from social media, similar to Australia

🚔 French police raid X's office in Paris

👥 Google's Gemini app is used by 750 million people, every month

🎥 Amazon will start testing AI tools for film and TV production

This startup, which is based in California, wants to create its own “operating system” for aircraft and automate their flight computers. It’s called SkyOS and, while a pilot is still needed, their technology has been able to handle the trickiest and most dangerous aspects of flying - from takeoff to landing.

The company has just raised over $300 million in Series C funding, which values it at $1.15 billion. It has been backed by several investors - including Fidelity and Qatar - and is now approaching the end of a lengthy FAA certification process for its flight control system.

Their automated system was initially developed for helicopters, which are arguably the most unforgiving aircraft to operate, but their ambition is far broader.

SkyOS has already been integrated on Black Hawk helicopters for the US military, and contracts are in place with United Rotorcraft, Air Methods, and Mitsubishi Corporation to roll it out across various rotorcraft and fixed-wing platforms.

Last year, the FAA granted final design approval for its flight computers - so the company now needs to clear formal flight testing to reach full certification. If they’re able to do that, it’ll unlock a huge number of opportunities in both the commercial and defence sectors.

They’re certainly one to watch in the coming year.

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.

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.