Harry Winser.com

So it turns out Meta’s Muse AI has a pretty serious zero-day exploit. As reported by Dan Goodin in Ars Technica, it appears the macOS app has an easy way to gain access to a users Muse AI account:

The zero-day allows any app or terminal command to gain access to the token that authenticates users to their Muse account. Meta developers designed the assistant so that any locally installed app or executed code, regardless of the macOS permissions it has, can change a long list of undocumented settings. Most of them are fairly innocuous, such as controlling dark mode. One setting, however, is anything but innocuous. It allows processes to change the endpoint where transcription occurs. Normally, it’s a server address operated by Meta. Attackers can exploit this flaw by changing the location to their own endpoint. Once that happens, the attackers have the token that gives complete control over the Muse account.
- Dan Goodin

Once an app has access to the account (by pointing the endpoint at the local app and stealing the token), it can ask Muse to perform tasks like writing malicious files to disk, record via the microphone, and even take photos. Yikes!

It seems Meta’s long history of questionable security and privacy continues…

I think the overhaul of Siri in iOS 27 is a big deal, enough to warrant such a title. Over the last four days,f I’ve found myself drifting to use Siri for all of my general, if sometimes silly, questions. Questions like “How much would it cost me to run a laser quest business?” (£50,000), and “What’s the purpose of that extra little pocket on my jeans?” (it’s for pocket watches). I no longer use Claude or ChatGPT. Siri’s chat app, while simple, delivers exactly what I need: a window where I can ask questions, get quick results, and move on.

I think ChatGPT has most to lose now that Siri is functional. Android has Google Gemini pretty much baked in (or Bixby if you’re really embedded in Samsung), and Siri now serves that purpose for Apple. You no longer need to download a separate app just to ask how to boil an egg. Apple has access to all of your personal information, going back years: texts, emails, calendar, reminders, the lot! And people trust Apple with their privacy. For the most part, users are ok with Apple indexing their data. Privacy has become a much bigger deal in the last five years; Meta, Google, and OpenAI have either destroyed, eroded, or not earned the same level of trust that Apple has. And Apple now gets to reap the benefits of it *.

ChatGPT has forever touted users and install counts, but it’s also pretty clear these chat apps aren’t sticky. None of them manage to lock users in like Instagram or email. Most of ChatGPT’s users are on the free tier, which means for each query someone makes, they’re burning money. Apple, meanwhile, has a mountain of cash, and has priced Siri usage into the hardware. Apple (and Google) just have to wait ChatGPT out.

Looking ahead, I would expect OpenAI to push even harder towards paying business customers. They need revenue to support their free user base. But if I were an OpenAI investor, I’d be looking down at my newly updated iPhone and questioning my financial choices.

* I think there’s a whole blog post about how Apple is only five years away from selling all our data, but this isn’t that piece.

This week marks the Nintendo GameCube’s 25th* birthday! It was the first games console I fully owned, and it’s the one I probably have the fondest memories of. I was the only one of my friends to have one, and loved it until the Wii replaced it a number of years later. I’d spend hours playing The Legend of Zelda: The Wind Waker, Harvest Moon: A Wonderful Life, and Star Fox Adventures.

Looking back, the GameCube made little sense. It had more graphical muscle than the PS2 yet didn’t end up getting the “big” games of its generation (such as GTA: San Andreas). It had a carrying handle, yet was a home console (were we supposed to pack it up and its cables to go on holiday with it?) It had tiny 8cm mini-DVD discs while CD/DVD became the standard for other consoles.

With all this in mind, it’s not surprising the confused little cube didn’t win out. It was a device with one foot in the past (pushing polygons equals customers), and one foot in the future (releasing games that were weird and fun). Meanwhile the industry experienced a shift to an edgier style, as gaming went through its awkward “teen” years.

Since its release though, it’s developed a bit of a cult following. The Legend of Zelda: The Wind Waker is still highly regarded for its art direction and style, while Metroid Prime nailed the dark and isolating environments well before the so called “adult” gaming machines did. Nintendo have been re-releasing some classics via their Nintendo Switch Online subscription, and while I’ve not yet jumped ship to a Switch 2 (a requirement to play these titles) with every new addition it’s making me itch to do so.

So I salute you, little GameCube; while you didn’t wow others when you were around, you certainly wowed me.

* Actually, that’s not strictly true if you’re European; it’s 24.

On Wednesday Apple unveiled the industry’s worst-kept secret - the iPhone Duo. It’s Apple’s first attempt at folding tech. I went into the “Surprise and Shine” event not feeling overly enthused, but their advert demo and subsequent hands-on have now got my attention.

I’ve long believed that foldables are the future of phones, but I’ve always understood why there’s been such resistance. I previously owned a Samsung Galaxy Z Fold 3 and loved it. However, a few things made it a device geared more towards “tech nerds” and early adopters.

First was the price; at over £1,000 it priced out casual users from giving the form factor a try. I myself bought it second-hand two years after release for half that. Second was durability; there have always been concerns as to how well the screens and hinge hold up over time. Mine died after its first year of use (luckily it was in warranty). Finally, the software was a bit undercooked. At best, the inner screen made some apps work brilliantly. Outlook comes to mind where it would display my email on the right while displaying my inbox on the left - very similar to a desktop experience. At worst an app would render in the middle of the screen with large borders (Instagram) or a menu bar would be cut off making the app borderline unusable. The front screen deserves an honourable mention; at an aspect ratio of ~24.5:9, no app looked good.

What we’re seeing with iPhone Duo is Apple’s attempt to deal with the device quality and software. They seem to be confident in the durability of the device (though, we’ll have to wait for full reviews before confirmation), as well as software support. With the software, there are some genuinely clever ideas such as the way the screen blurs while you’re opening it (this has almost sold me on one alone). The Netflix demo also looks pretty neat.

The price at £2,000, while in line with similar devices from the likes of Samsung, is still going to alienate the casual user.

All this is to say that I’m now sorely tempted by the Duo. Which is exactly what Apple wants me to think. Clearly they did “Surprise” me though I’m not sure where my “Shine” is yet.

Last week OpenAI’s president Greg Brockman made a pretty wild claim when announcing OpenAI’s latest model GPT-6 Astra:

“I think that if we fast-forward a couple of years, when we look back and say, ‘When was it really that AGI was created?’ I think it’s going to be about this time, and I think it might be about this model.”

- Wired

No, we have not reached AGI yet. Nor does it seem likely in the next few years. LLM tools are certainly getting better at automated tasks, software development, and document writing, but they’re still heavily reliant on human interaction and direction. I won’t believe OpenAI or Anthropic or any other AI company until they “hire” their own AI to a position on their C-suite.

The term “AGI” means next to nothing. Altman said so himself last year. “AI” and “AGI” aren’t well-defined terms*, which gives AI companies excellent wiggle room for marketing while also avoiding legal liability when their tech doesn’t work as expected.

We must remember, as OpenAI is looking to IPO soon, they’re going to say stuff like this. OpenAI has to hype their models to pump their share price as much as possible while the going is good, and make off with the big $ sack before the circular deals and crazy accounting catch up with them.

* humans have had centuries to think about this topic, and produced thousands of books (and podcasts) discussing what even is “intelligence”.