Midjourney is an AI image generator and copyright violation machine. What’s the weirdest and dumbest business pivot an image generator company could possibly make?

Yesterday, Midjourney CEO David Holz announced Midjourney Medical! No, they didn’t call it “Medjourney.” They’re going into medical imaging! Ultrasound full-body scanning! [Midjourney]

They call this “ultrasonic CT”. In Midjourney’s vision, you’ll be lowered into a pool of water. Ultrasonic sensors will scan your whole body over sixty seconds. The system will compute a 3-D scan of you. Midjourney describes the process in more flowery terms:

It starts by stepping into a shallow pool of golden light. You then begin to descend into the water. Your body passes through a ring of underwater sensors, each acting like a dolphin, using its echolocation.

Actual medical advances don’t talk about “golden light” or compare themselves to dolphins. At absolute best, Midjourney is selling a vibes-based “wellness” experience to rich new age suckers.

There seems to be a prototype scanner in existence — Holz says they’ve scanned maybe a dozen people.

But Midjourney has fabulous plans for the future! The “Midjourney Spa” will apparently open in San Francisco at the end of 2027: [Midjourney]

It will have hot tubs, saunas, cold plunges and 10 scanners with the capability to do more body scans a year than all MRI scanners on Earth combined.

Midjourney is careful in their wording. Their ultrasound scanner is not an MRI — they just compare it to MRI a whole lot. David Holz says “We’ve dreamed of something as powerful as MRI.”

Midjourney plans to deploy 50,000 scanners around the world in the next six years!

So there’s a few problems with Midjourney’s scanner idea.

Ultrasound scanning doesn’t penetrate the whole body. It doesn’t work through bone or air. And there’s a lot of both of those in your body. Ultrasound just doesn’t do full body. Ultrasound, CAT scans, and MRI are different scans that see different things.

And if you call something medical, there’s still actual regulations on that: [Verge]

various medical applications would require FDA clearances, but for now, Midjourney Medical says it’s working on “body composition maps” that don’t require the same level of clearance as diagnostic imaging.

The good news is, this thing does not use generative AI. Holz told the press conference: [Bloomberg]

We’re not even using any AI in this yet — just really cool hardware and software.

So you won’t be asking if it scanned six fingers or seven.

Is Midjourney’s scanner medically better than nothing? Well, not really — it barely exists. This scanner is not any sort of medical product, it’s a completely experimental toy.

None of this has been tested. None of the stuff Midjourney is talking up has been verified by anyone.

This is Theranos-level vaporware. Every radiographer who’s heard about this thing has called it out — because they’ve all heard this sorta napkin-scribble idea from idiots before.

There have been other AI medical startups reminiscent of this. Forward talked up plans for AI-powered CarePods — drop a patient in, do automatic checkups! The CarePods did not work — they didn’t do a single thing. Forward soaked up venture capital cash and finally went broke at the end of 2024. [Fierce]

Midjourney has not accepted venture funding in its history. This ultrasound plan makes me wonder if the AI copyright violation business isn’t going so great, and they’re about to start calling round the VCs.

Can Midjourney’s ultrasound scanner become a good and useful medical device in the future? It’s hard to prove nothing like it could ever happen.

But I think we can say that the guys selling you golden light and dolphins and a spa experience aren’t the hot favourites.

• Video — Podcast

Unreal Engine is a video game engine produced by Epic Games. It’s what Fortnite is written on top of.

Epic CEO Tim Sweeney is convinced the slop machine is the future. In November, he was claiming the Steam store — Epic’s direct competitor — should stop tagging games using AI, because “AI will be involved in nearly all future production.” Uh-huh. [Twitter, archive]

Sweeney is working to make his fabulous future vision a reality. Epic’s announced an early preview of Unreal Engine 6, the next version, coming in late 2027.

Guess what it’s got? AI! Even better, Tim’s still trying to make the Metaverse happen! Inside Unreal Engine!

Here’s the announcement, from dev lead Marcus Wassmer: [Epic]

We’re building development pipeline features such as an MCP with integrations for Claude, Gemini, and others.

Actual game developers — not the studio executives — are not tempted. The devs hate using AI and think it’s trash.

Austin Wood at Games Radar tried to do a both-sides piece on AI in gaming. He even tried to claim there were game developers — not just executives — who “do see value in AI.”  But he couldn’t produce a single one: [Games Radar]

Perhaps pro-AI developers didn’t want to talk to me or I just didn’t run into any during my survey, because I heard an overwhelmingly negative assessment of generative AI’s origins, capabilities, and risks. By the end, I’d heard dozens of developers make a case against using gen AI at all.

So what does Epic think the game devs will use the slop bot for?

as creativity and productivity multipliers so that teams can focus their efforts on the essential creative and technical tasks of development rather than time on time-consuming manual tasks.

That means the game assets. Images, backgrounds, sounds, and so on.

The problem is that’s not a disposable part. Games aren’t just a framework you drape arbitrary slop over. How it looks is part of how the game feels! It’s load bearing!

The paying customers sure think so. When Call of Duty: Black Ops 7 came out last November, it had a pile of returns — specifically because the sloppified game play looked so much worse than the non-AI trailers.

Unreal Engine 6 will itself be written with AI slop code. Should do wonders for reliability if you try to use it.

But it gets better. Remember, from 2021, the ridiculous promise from crypto bros that NFTs would mean you could swap hats and guns between unrelated games? Sweeney still wants this. From the announcement:

This means you’ll have the option to use a player’s entitled Fortnite Outfits in your own games, and you’ll get the tools to build Outfits for your own games that work inside Fortnite.

… We see this as the first step toward building a shared economy for smart assets: functional assets with logic and functionality that work across games

Epic wants to bring back NFTs!

Sweeney also wants to build a game chat network into Unreal Engine itself, I guess because Discord is getting money that he thinks he should be getting: [Inven Global]

the ‘cross-game social link’ feature we are trying to implement through UE6 will have tremendous business value. Users will be able to form parties via voice chatting with friends in Game B while connected to Game A, and encourage them to try a new game.

This network won’t be any equal sort of thing — it’ll be firmly gated by Epic. Sweeney wants you, the outside developer, to put Fortnite hats into your game so Fortnite can use them. The other way? Ehh, good luck with that.

Now, you might think the AI vendors are already running out of all the money in the world, and they’re putting up prices as far as they think they can get away with. I’m pretty sure 2027 will be hilarious for Unreal Engine with AI. And expensive.

In the meantime, start learning the Godot engine. Which doesn’t accept AI code. [Godot]

• Video — Podcast

Chatbots are a security hole the moment you let them do real work — because you can always prompt-inject a chatbot.

And if Microsoft does this stupid thing, you can break in and get confidential company information.

This one comes from Varonis, who found the “reprompt” attack in January — where you hack Copilot by asking it twice.

Varonis’ new hole is called SearchLeak. It’s a one-click attack — the victim has to click on your link. But it’s stupidly simple: [Varonis]

1. The attacker sends an email with a link in it.
2. The link calls Copilot’s enterprise search with a search string that’s actually a prompt.
3. The prompt tells Copilot to call a web address controlled by the attacker, with the victim’s data encoded in the address.

The web address is sent to the Bing search engine — which Copilot trusts. Bing calls the address! And your data is leaked.

This attack chains three small holes into an exploit:

1. If you search in enterprise Copilot, the search string gets passed directly to Copilot as a prompt — it takes it as instructions.
2. There’s a race condition — an image tag in the chatbot’s response loads before the output sanitiser can kick in.
3. The bit where Bing calls the attacker’s web address (that image tag) for you.

Microsoft marked this a critical issue and patched it on 4 June. [Microsoft]

But using a chatbot for real work is unsecurable by its nature. As well as an extremely stupid thing to do. So I’d expect more of these.

• Video — Podcast

Last September, programmer Mike Judge asked: if AI’s so good for coding, where are all the little AI apps? Where are the widgets someone just vibed? Where’s all the AI shovelware? He worked out the numbers — and there had been no uptick in apps going to the Apple or Google stores.

This year, the AI bros are finally pumping out the little apps! And guess what? Nobody wants them.

There’s a new paper from NBER: “Writing code vs. shipping code: Productivity effects across generations of AI coding tools”. [NBER, PDF]

Their thesis is that AI has led to tremendous “productivity gains” in coding — but not in the resulting apps being used.

Their idea of “productivity gains” is just lines of code. Churning out chatbot code is productivity now. They spend the first forty pages of the paper analysing the process of vibing more lines of code.

So that’s great, we have a garbage fountain, very productive. But — does anyone care? That question is covered quickly in the last nine pages of the paper:

Despite this expansion in supply, we find that total app usage within the first three months of launch has not increased in any of the four marketplaces.

… The share of new applications that fail to reach even a modest audience has risen across marketplaces, suggesting that the supply-side expansion is concentrated in applications with little to no user base.

That is: the users do not care about the flood of crappy AI spam apps. Why should they?

The paper doesn’t establish that the garbage flood is definitely AI-coded apps. But they think that’s the obvious source for the sudden increase in new apps, and I concur.

There’s a lot of wannabe developers who suddenly feel empowered by the vibe code. And the graphs in the paper show an uptick in new apps in what the authors call the “agentic-coding era.”

Why don’t the users care? The authors posit that the users just need to catch up and get with the programme:

the usage response may simply be slow, since discovery and adoption take time

Or maybe the users don’t care about trash apps.

If you look at the graphs on page 42, the number of apps is going up, up, up!

But the total usage of all apps is flat — or declining. Adding more apps does not mean more apps getting used.

And then there’s the ratings and downloads. Nobody rates or downloads the flood of vibe garbage. It’s spam and nobody cares.

The authors of the paper are AI boosters who seem really disappointed that nobody cares about all this slop code. They even misdescribe their own data in the paper’s abstract:

Large task-level AI productivity gains have therefore translated only partially into shipped and used software thus far.

If you call zero or negative a sort of “partially.” Because zero or negative is what their own graphs show.

The authors keep calling wheel-spinning busywork “productivity” — and not a complete waste of effort, money, and everyone’s time.

• Video — Podcast