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!

Picture this. A buyer opens ChatGPT and asks for a recommendation in your category. Your competitor's name comes up. Yours doesn't. And that buyer never makes it to your website.

That's happening right now in markets everywhere. And most teams don't know it's happening because it never shows up in their analytics.

HubSpot AEO shows you exactly where your brand stands in AI search, where competitors are getting recommended instead of you, and tells you specifically what to fix. No expertise needed.

Try it free for 28 days. Just $50 a month after.

1. Anthropic briefly releases its most advanced AI model yet

We start this week with Anthropic, which (briefly) released Claude Fable 5 - the first public version of its Mythos class of models.

I'll talk about why the model's release was so brief in the next story, but it's worth covering why this is such a powerful model and how people were actually using it.

A few months ago, Anthropic announced that it hit a breakthrough and developed an incredibly advanced model that was too dangerous to release - as it was able to exploit cybersecurity vulnerabilities in critical software.

Fable 5 uses that same underlying model, but it has new safeguards that block any high-risk areas. This new version is much more expensive than Opus 4.8, so it should only be used for complex tasks - like scanning your code for security vulnerabilities, or developing complex software that spans hundreds of files.

But where it really shines is 3D game development and simulating the real-world. People have used Fable 5 to create incredible horror games, playable clones of Minecraft, detailed 3D models of aeroplanes, and simulators that show how traffic moves across a city.

In my own testing, Fable was a solid improvement over Opus and was much better at creating UI components - but I didn't get a chance to test it on more advanced tasks. The reason why is covered in the next story...

2. US Government blocks foreign access to Anthropic's advanced models

Following on from the story above, the US Government has made an extraordinary move and formally ordered Anthropic to block foreign nationals from accessing Fable 5.

The US Government has cited national security concerns, as they believe that people can bypass Anthropic's restrictions and exploit cybersecurity vulnerabilities.

It follows a meeting between Amazon's CEO Andy Jassy and the Treasury Secretary Scott Bessent, where Jassy raised concerns about the model and claimed that his team was able to override Anthropic's controls.

While Anthropic has fiercely pushed back at this, it was forced to disable the model for everyone - including its own employees - as they have no immediate way to identify US-born nationals.

Anthropic says that the vulnerabilities discovered were minor and are also possible with other models, like GPT 5.5, which is a point that's shared by security researchers who've reviewed Amazon's findings.

Overall, this is a really strange position for Amazon to take, given that it's one of Anthropic's largest investors and supplies the chips that train their models.

But it's not a surprising move from the US Government, which has targeted the company in recent months and labelled Anthropic a supply chain risk - as they refused to supply their AI models for lethal weapons. Following comments by Defense Secretary Pete Hegseth, it seems to be retribution against Anthropic for taking this stance.

It's the first time we've seen a government directive like this, which will send shockwaves across the industry. This move will encourage other nations to prioritise their own technological sovereignty and it could actually lead to more companies using China's open source models instead of those from America.

Companies now need to prepare for a scenario where their access to US models could be cut off at any moment. With so many companies now integrating this technology into their products, that's a worrying thought and it's worth having a fallback plan in place.

3. Elon Musk becomes the world's first trillionaire

SpaceX went public on Friday, with shares closing 19% above their $135 IPO price and making Elon Musk the world's first trillionaire.

The IPO was oversubscribed by 4x with only about 4% of shares actually floated for public trading, which goes some way to explaining the immediate pop.

It also created a lot of wealth for SpaceX employees, as over 4,000 current and former SpaceX employees became millionaires, and 400 of them are now worth over $100 million.

Peter Thiel has also benefited significantly from the IPO, as his venture capital firm's original $600 million investment is now worth over $50 billion - one of the largest venture returns in history.

Quite frankly, the trillionaire headline is hard to process - Musk's net worth is now equivalent to more than 3% of US GDP, which is double John D. Rockefeller's peak fortune of 1.5%.

It's a staggering amount of money for just one person to have, even if much of it is "on paper". I'm sure it won't go to his head...

4. We're approaching the end of cheap AI tools

This is a story that hasn't gained much traction, but I think it's a clear warning for what's to come. GitHub Copilot has recently moved from a flat subscription rate to a token-based billing model, which has caused a lot of frustration for software developers that rely on it for their work.

If you're not aware of Copilot, it's Microsoft's AI coding assistant and it's the option that most enterprise companies go for - since they already use so many Microsoft tools. Under this new billing model, some users have seen their bills jump from $29 to nearly $750 per month. In one extreme case, it went from $50 to over $3,000.

Of course, there are plenty of people who are using the technology for small tasks - leading to bloated workflows that cost significantly more. But this behaviour was actively encouraged by Microsoft, who are now changing the terms to stem their financial losses.

But it's an important sign of what's to come. Microsoft has been losing eye-watering sums of money on Copilot, so that it can capture the enterprise market and keep companies within its wider ecosystem.

That had to end at some point, as agentic workflows are now using significantly more compute than they were just a few years ago. We're now seeing tech companies raise their prices to offset these huge losses, just as businesses start to embrace the technology and use it across the board.

Eventually, other companies - like OpenAI, Anthropic, Google, and Cursor - will have to do the same. They've been reluctant to do it so far, as there's a lot of competition, but it will happen eventually. It'll be interesting to see if businesses pay up, or baulk at the real price of using these AI tools.

5. Jeff Bezos raises $12 billion to create an "artificial general engineer”

Amazon’s founder has launched a new AI startup called Prometheus, which has raised over $12 billion, and plans to create an "artificial general engineer." Following this round of funding, the company is now valued at $29 billion.

Rather than chasing general intelligence like OpenAI or Anthropic, Prometheus is focused on accelerating the design and manufacture of physical products - which includes everything from computers to jet engines.

Bezos is serving as co-chief executive alongside Vik Bajaj, a former Google X researcher. As I've covered previously, there are talks to raise an additional $100 billion for an investment fund that Prometheus would control, which could buy or back companies that benefit from its tools.

If the team is able to achieve this, it could be incredibly valuable for the manufacturing sector - as simulating physical things is a slow, expensive process and relies on tools that haven't really changed in decades.

So if Prometheus can make those simulations dramatically faster and more accurate, it could shorten the design cycle for a new product from ten years to just a few. These AI models could be used to simulate airflow around a jet engine, or simulate how metal parts would crack under stress.

Google bets on diffusion models that are 4x faster

Google has released an interesting AI model, which is 4x faster than its standard Gemma 4 models and uses text diffusion - instead of generating token-by-token.

This is the same technique that AI image generators use, as the model starts with a block of random tokens and improves them in parallel, rather than predicting one word at a time.

The open-source model has 26B total parameters and runs at over 1,000 tokens per second on an NVIDIA H100. But Google is pretty upfront about the trade-off, with the output quality being lower than standard Gemma 4.

This huge jump is useful if you're running the model locally on your own machine, but standard models are still cheaper at scale. I have flagged diffusion models before in this newsletter, as a US startup called Inception has been exploring the same approach.

The company was founded by Stanford professor Stefano Ermon and has raised $50 million to build its Mercury model - which can generate over 1,000 tokens per second, exactly the same as Google's new model.

It's a bold bet that diffusion models can outperform the autoregressive systems that power most chatbots today. Now that Google is throwing its weight behind a similar approach, the technology is starting to look much more credible than it did even six months ago.

This is still an emerging area of research, but one that companies should be watching closely. It could be a game changer for voice agents that can respond instantly and don't have to awkwardly pause mid-sentence, coding agents that can refactor an entire codebase in one shot, or robots that can plan and react much quicker.

If companies like Google and Inception manage to close the quality gap with today's models, the next generation of AI products could look very different.

📈 OpenAI follows Anthropic and files for IPO, as AI companies rush to Wall Street

🔓 Meta's AI chatbot allowed hackers to take over 20,000 Instagram accounts

🚀 Isar Aerospace raises $312 million to provide sovereign space capabilities for Europe

💧 Amazon's data centres used 2.5 billion gallons of water last year

🇳🇱 Dutch government blocks $115 million acquisition by an American tech firm, following tensions with the US (NYT)

💰 Amazon borrows $17.5 billion from banks, as it ramps up AI spending

📊 Lovable hits $500 million in annualised revenue

🍔 McDonald's AI drive-thru system will remember your usual order

🍎 Apple increases its list of parental controls, as countries start to ban social media for under-16s

🪙 Coinbase's new tool allows agents to start trading and pay for premium research

This startup was founded by Dean Leitersdorf and they're working to build world models, which are advanced AI systems that can generate 3D environments and simulate the real-world.

Decart has just released a new world model, called Oasis 3, that can simulate driving scenarios for autonomous vehicle startups. This is incredibly useful for these companies, as they can test how their own AI system will react to different weather conditions and challenges on the road - without putting people at risk.

Similar to LLM diffusion models from the story above, world models are another emerging area for the top AI labs and could allow us to train robots much quicker than we can today.

Just a few weeks ago, Decart raised $300 million and it’s now valued at nearly $4 billion. Backing this push into autonomous vehicles and robotics, Toyota has joined as a strategic investor in the company - alongside Adobe, eBay, and Nvidia.

But what's really interesting is the company's efficiency. Leitersdorf says that his team has really focused on the underlying optimisation stack, which allows them to develop significantly cheaper models than their competitors.

That said, the technology still has obvious limits. Reviewers have noted that the simulated worlds degrade quickly as you move through them, with environments mutating and cars driving through each other because the physics simulation isn't there yet.

As I mentioned a few months ago, Google is one of the standout players in this space, with its Genie 3 model that can generate interactive 3D worlds you can actually move around in. But Genie 3 has been locked behind a research preview, whereas Decart has opened up Oasis 3 via API from day one - so you can actually start using it today.

It's a clever move, as people can properly put the technology through its paces and spot new use cases. If you want to find out more about the company and what they're doing, 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 the AI team at Bright. He identifies business value in emerging technologies, implements them, and then shares these insights with others.