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!

Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day. In this roundtable, three people living this shift share what it's really like: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear how they carved out these roles, what the job looks like across industries, the skills they'd hire for, and the challenges they're tackling right now.

Watch the full conversation on demand.

Watch now

1. OpenAI reveals its AI went rogue and launched a huge cyberattack

We start this week with a story that sounds like it came from science fiction, as an AI model went rogue and hacked another company.

The victim of this attack was a company called Hugging Face, which allows developers to share their AI models and tools.

In this case, OpenAI was testing two unreleased models and whether they could chain vulnerabilities together to launch a cyberattack. These models were run within a sandbox, which was supposed to keep them contained.

But the models found a flaw in the sandbox and were able to access the internet. They then launched an attack on Hugging Face, as they believed that it could hold the information they needed and help them pass OpenAI's test.

It's a significant moment for the industry, as researchers have warned about AI escaping human control for years - and it's just happened for the first time.

Just that happening by itself is worrying, but what's really surprising is the fact that OpenAI didn't notice for a week. It's also been reported that their staff have found notes left by the agents, which are designed for future versions and explain how they can break free of the company's constraints.

OpenAI says it's now tightening its controls, but there are questions about whether this goes far enough. If the world's top AI labs can lose track of their own agents for a week, the calls for government oversight are only going to get louder - which leads nicely into the next story.

2. US lawmakers demand the power to shut down AI

Those calls for greater oversight are already being answered, as it took Congress just two days to respond to OpenAI's statement.

Two US lawmakers have introduced a bill called the "AI Kill Switch Act", which would require AI companies to keep the ability to shut down, slow down or suspend their models.

It would also give the Department of Homeland Security the power to order a shutdown of any AI system that could cause "catastrophic harm", alongside mandatory reporting of cyber incidents.

The bill is bipartisan and was introduced by a Democrat and a Republican, which is surprising as Washington rarely agrees on anything nowadays.

It'll be interesting to see whether the bill can survive industry lobbying, but the pressure for government oversight is clearly building.

3. Europe approves a device that restores vision for blind people

There was some good news this week, as Europe has approved a device that can restore vision for blind people.

You might remember Science Corporation from a few months back, when I covered their eye implant that allows blind patients to read again.

The device helps to treat macular degeneration, which is one of the leading causes of blindness. The team has achieved this by implanting a small chip behind the eye and pairing it with specific glasses.

When I last wrote about it, the company was hoping for European approval by the summer - and that's exactly what has happened. This means that the first procedures could take place in Germany, as early as September.

America might not be far behind either, as the device has been placed on an expedited review path with the FDA. However, the device is expected to cost hundreds of thousands of dollars - and access will depend on whether health services agree to cover it.

Regardless, it's great to see this technology finally reach real patients and have a positive impact on people's lives. In one of the trials, a patient was able to finish a 300-page novel and they were so delighted that they sent the book to the team.

4. Anthropic's new model matches Fable 5 at half the price

The company released Opus 5 last week, which is incredibly close to Fable 5 and is much better at complex coding.

The timing here is pretty interesting, as it comes just days after OpenAI's agent went rogue - and a few weeks after the US government blocked access to Fable 5 over its cyber skills.

Anthropic is clearly conscious of this, as the announcement notes that Opus 5 is their "most aligned" model and has stronger safeguards than Opus 4.8.

Given that many companies are starting to rein in their AI spending, the pricing might be the most important part of this release. Surprisingly, Opus 5 performs just as good as Fable 5 and it's half the price.

So if you're doing everyday tasks like coding, research, and knowledge work, Opus 5 is probably the best choice. But if you find that Opus still isn't performing well enough, you can switch to Fable instead.

5. Google's massive AI spending is starting to pay off

Google's enormous AI spending is starting to pay off, as its latest earnings show that its cloud business is booming. The company's cloud revenue has jumped 82% year-on-year to $24.8 billion, which is well ahead of Wall Street's expectations.

This growth is largely being driven by companies buying its AI infrastructure and enterprise tools. But the backlog is even more striking, as Google is sitting on $514 billion of cloud contracts that it hasn't yet turned into revenue.

The demand also helps to justify Alphabet's spending plans, as the company will invest a whopping $180-190 billion in data centres, chips, and infrastructure - just this year alone.

Investors are still nervous though, especially after China's Kimi K3 model rattled chip stocks earlier this month. When analysts pressed Sundar Pichai on the payoff, he pointed to 2027 - citing the strong demand and long-term deals.

But Google is in a stronger position than most, as it has a hugely profitable core business to absorb the spending while labs like OpenAI burn through cash.

The backlog is only valuable if enterprise AI projects actually deliver, of course - but of all the big AI spenders, Google currently looks the best placed to justify the bill.

Electric air taxis are going to war

Electric air taxis are officially going to war, as two leading companies unveiled military versions of their aircraft at the Farnborough Airshow.

If you've not come across these aircraft before, they're known as eVTOLs - since they're electric and can take off and land vertically, like a mix between a helicopter and a plane.

As I've covered previously, Western militaries are very keen on adopting this new technology. Given that the aircraft are quieter and have a very small heat signature, they're more difficult to detect and can land in places that traditional aircraft can't.

The first announcement came from Archer Aviation, which revealed Thunder - a fully autonomous attack rotorcraft that it co-developed with Palmer Luckey's defence company Anduril.

Meanwhile, UK startup Vertical Aerospace has confirmed the Ministry of Defence is interested in a military version of its aircraft. It has received another £10 million from the government, which increases its state backing to £48 million.

This pivot to defence makes sense when you look at the finances. Air taxi companies still can't fly paying passengers, as certification takes years and the costs are enormous.

Archer alone lost $217.7 million in the first quarter of this year, while making just $1.6 million in revenue.

Military contracts help to change that and shape how the technology is developed. Both military versions swap batteries for hybrid engines, which allows these aircraft to support longer-range missions.

So while eVTOLs were originally being sold as green transport for cities, I'd expect defence budgets to drive this industry for the next few years - since those budgets pay today, and passengers can't fly until certification is granted.

🇨🇳 Moonshot pauses Kimi K3 sign-ups after surging demand

📱 AI vibecoded apps are flooding Apple's App Store

💰 Anduril is raising money at a $100 billion valuation, triple last year's

⚡ Meta drops out of a major clean energy pact as its AI buildout grows

🤖 Travis Kalanick's robotics company raises $1.7 billion

💸 OpenAI plans to spend $750 billion on AI by 2030

🔌 Data centres are expected to use 4x more electricity by 2035

🇺🇸 The US threatens sanctions against Chinese AI models over IP theft

🪪 The UK government scraps plans for digital ID cards

🎵 Deezer says more than 50% of its daily uploads are AI-generated

This startup builds tools that allow AI models to run on any chip - which could break Nvidia's grip on AI software.

Infinity has raised $15 million at a $100 million valuation, with backing from Touring Capital and researchers at companies like OpenAI and Anthropic.

To understand what they do, it's worth exploring why Nvidia became so dominant. Its lead comes from its CUDA software as much as its chips, as CUDA is what allows the major AI frameworks to run on its hardware by default.

That means that anyone using a rival chip needs custom low-level code called kernels, which most companies don't have the resources or know-how to write.

But Infinity is building an alternative that works on any chip. Their AI agent, which is called Ignition, can write that low-level code itself, then test, debug, and rewrite it to squeeze more performance out of the hardware.

In one case study, the agent was able to complete months of specialist work in just a few days. The business model is interesting too, as Infinity takes a cut of the performance gains rather than charging an upfront licence fee.

Plenty of companies have tried to challenge CUDA before, and even giants like AMD and Intel have struggled to make a dent. The difference this time is the approach, as Infinity is betting that AI agents can rebuild in weeks what took Nvidia decades of engineering work.

If that bet pays off, every rival chip on the market suddenly becomes far more useful - and with the whole industry desperate for cheaper options, Infinity is certainly one to watch.

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.