For context, there have been issues with MacOS 26 which have led many people to defer upgrading until MacOS 27 is available, and MacOS 15 is the previous version.
Also a liquid glass hater, but for what it's worth, 27 is supposed to make it a little bit better. I would prefer to go back to what I had before I had to upgrade into this horrible UI, but this is better than nothing.
I've been running the 27 beta and it fixes so much of the ugliness I hated in Tahoe. Lots of places had toolbars restored, solid areas that indicate where the window can be grabbed. The shitty icon spam in menus is back to sanity. And window corner radii are consistent and less bulbous.
macOS 17 is decisively faster on my MacBook Air M1 16 GB than macOS 26/27 is on my MacBook Pro M2 Max 32 GB. And I don’t mean just the animations themselves but lag/sluggishness/responsiveness in general.
You can already turn off most glass effects in 26 via "System Settings => Accessibility => Display => Reduce Transparency".
I had that turned off years ago (for reasons I don't remember), and was wondering what all the fuzz was about when 26 came out because I didn't see much of a difference ;)
IMHO the actual important visual changes in the 27 beta is that rolls back the bizarre oversized corner radius in Finder windows, and they also got rid of the 'every menu item must have an icon' idea.
Also on the beta. Agreed. Definite visual and UX improvement over 26, fixing the most egregious issues with 26 implementation of Liquid Glass, but maybe not quite as good as 25 overall.
As long as the UI improves and I can ignore all the AI stuff they're starting to push through, that's fine with me, though like many longtime macOS users, I'm not holding my breath for a bug-free experience.
A couple of weeks ago, I had a Baltimore Oriole (a cool-looking bird, not a baseball player) in my yard. They aren't rare, per se, but they are uncommon.
Took my iPhone out to snap a picture, and pressed the camera button. I hadn't used it, since upgrading to 26.
It takes the picture. It's there. I can see it, but it won't let me save it. Instead, it wants to tell me about the cool new voice-activated AI retouch feature. There was no way to save.
I probably could have figured it out, but I was so furious, I just nuked the picture.
For those of us older than just 10 years of using macOS, the older Apple OSes have instilled within us the desire to never install the X.0 release and wait until at least the X.1 release. The bug-free experience is a myth
I found if you turn off transparency effects and turn on high contrast in the accessibility settings, it's actually a pretty nice looking UI. Still has some dumb quirks like inconsistent corner radiuses and text appearing under UI elements (because they're assumed to be transparent, but aren't). But all software is trash in 2026 so be thankful it isn't even worse, I guess.
Been doing that for years now already (performance boost). You can also do high contrast mode per app (if were talking about iOS). Must have for iMessage I think personally as it makes hard to read Green Bubbles a nice dark green that is very pleasant.
I bought a MBP mainly for the hardware, otherwise I'd have stuck with Linux. OS-wise, the jump from Monterey (which I'd last used) to Sequoia was smooth and still feels that way. I prefer not to notice the OS at all, something that I feel would be hard to do on Tahoe due to all that liquid glass and dumb transparency and animations.
Fortunately, those can run Linux. I recently installed Arch on an Intel T2. The only issue is that it does not have a TPM module, so the LUKS password needs to be manually entered at boot.
Love to see this, now if only they could revert the whole shimmering oil slick disaster that is "glass" on iOS (unlike macOS, even if I "reduce transparency", the oil slick effect still appears in many places, in addition to all the spacing issues).
Same. It's not the worst thing in the world, at least with Reduce Transparency enabled in the Accessibility settings, but I still don't feel any inclination to upgrade. My personal Mac Studio is macOS 15 and my work MacBook is on macOS 26, and I don't think there's a single thing that I find to be better on the work laptop than on my personal machine.
I might update to macOS 26 in September to be ready to update to macOS 27. Being two versions behind doesn't seem reasonable and I'd rather be on the "Tahoe but less shitty" version than Tahoe itself.
Same thing happens almost every release. I've stopped updating my Mac machine until I see something in the release notes I literally have to have in order to continue doing macOS/iOS builds, otherwise I'm staying on the version I've validated to work, and I know the existing bugs with.
Yeah, I thought so too, but surprise surprise; some months ago one of the "minor" updates "broke" ("upgraded") something that made my CI/CD setup stop working, that's when I dropped the idea that Apple even do "minor" updates anymore.
26.0 had a very annoying video jitter issue, but that was the first things that I noticed to be fixed in the next 26 release. Other than that, it worked just fine.
Did they un-hardcode the corner radius? I mean were they able to? I mean not that that anyone at this point needs convincing how utterly disgusting incompetent they’re at software.
I was shocked to find out the inconsistent corner radii in Tahoe was an intentional design decision. I figured it's so obviously bad that it was just sloppy work. But one of the WWDC session videos bafflingly clarified that it depends on whether the app has a toolbar or not.
I am fairly convinced this was all designed only in graphic tools, and looked nice in static mockups, like those fake desktops with overlapping small windows widely spaced, etc. The design folks probably all had giant studio displays with plenty of room, etc.
Then the reality of "what about toolbars", "what about dark mode", "what about laptop screens", etc were all afterthoughts and resulted in bolt-on fixes like that.
Map the amount of fixes with "... improved bounds checking...", "...improved memory handling...", "...improved memory management..." into the amount of developer, QA and release management teams salaries per hour, versus other stuff they could be working on, and that gives an approximate value of how using specific languages maps into monetary loss, and why companies are starting to care nowadays, given computers are always exposed to the world network.
>>, and that gives an approximate value of how using specific languages maps into monetary loss, and why companies are starting to care nowadays, given computers are always exposed to the world network.
You need also factor development time and ease of finding developers willing to work in a specific language. There are other factors like readability of the code (very verbose languages are likely to be worse) and cost of maintenance - languages forcing a lot of abstractions are likely much worse.
If you just meant apple-targeting developers, then yeah; you're right that those languages are in decline. But if you meant developers in general, I think you'd be surprised how many growth sectors are hiring C programmers. They're often not SaaS tech companies, but they are massive and many are growing. Hardware, industrial control systems, defense/aerospace ... there's a ton there, the spaces in which they hire just don't overlap a ton with the spaces frequented by hacker news. Also, a lot of them aren't US based companies.
I hope that changes over time, since I definitely agree that the downsides of C-family languages massively outweigh the downsides of competitor languages.
Which is why WG14 and WG21 caring about security would be quite relevant, but alas, priorities.
C could have gotten slices already in the 90's, the concept already existed in other languages, and even Dennis Ritchie made a fat pointer proposal into that sense.
The others, let see if anything related to profiles actually gets into C++29.
Citation needed. I don't think there's a correlation there. Over-architected Java spaghetti is verbose and unmaintainable. Under-architected Perl code golf that metastisized is terse and unmaintainable.
> languages forcing a lot of abstractions are likely much worse
Citation needed. C++ has had some very high-level abstractions on top of a low-level runtime for awhile, and plenty of people have decided to use it and hire for it regardless. What counts as an "abstraction" or "forced abstraction" is a very very subjective topic.
Even C can be super abstracted, see early 1980-90's business software written in C, with nice stuff like Yourdon Structured Method, leading to over-architected C spaghetti with macros, a decade before Java came to be.
The problem is the lack of interest since Morris worm came to be, to provide better mechanisms in said languages, until governments and key big tech names decided it was time to change existing practices.
It was my vague understanding that by the time you implemented all the apis needed to run normal software on top of that, you either have enough apis that different tasks can still compromise each other, or you have shoved everything into a single task with very little isolation between normal user processes. In either case, it doesn't seem like you actually gained so much. What am I missing?
Lots of "in collaboration with Claude and Anthropic Research" mentions, no mentions of other labs. I'd assume Apple already had access to whatever the most powerful model is at the various US-based labs, but perhaps not?
Those were voluntary disclosures by two Anthropic researchers and the security firm Calif. I know one more CVE on the list that was discovered using an AI agent and wasn't disclosed as such. I suspect there are many more.
The gist is, OpenAI hired a high ranking Apple employee who helped other Apple employees get hired by OpenAI and exfiltrate Apple trade secrets in the process.
> Lots of "in collaboration with Claude and Anthropic Research" mentions
I wouldn't say 4 is lots. The entire list is massive. I haven't counted myself, but someone claimed that macOS 26.6 has the all-time record with 155 CVEs.
Considering that in Feb 2026 (https://support.apple.com/en-us/126348) Claude wasn't mentioned even once, 4 sure sounds like "lots" compared to nothing :) But you're right, it's subjective ultimately.
Do they? As in Claude but on premises? Wonder if this is gonna be the solution that e.g. banks will require, exactly like they do now for cloud services (e.g. Azure on premises).
Pretty extreme solution… you can get Claude models from AWS Bedrock and Google Model Zoo. These are both very helpful for compliance and security, but do require you to have a cloud strategy.
I don't believe it's published anywhere, but it's common knowledge to Apple engineers. I can second the poster's assertion that they run Claude internally.
I don’t know about hosting it internally but a an Apple focused podcast I listen to (Accidental Tech Podcast) has mentioned numerous times in the last few months they’re using Claude heavily. And they know Apple insiders.
This may be a naive take, so if anyone has insight please feel free to share, but across Windows, Mac, and Linux OS's I see many cases of path parsing vulnerabilities resulting in sandbox escapes, code execution, or data access issues. When presenting the user with a file picker or command-line input, is it really needed that the software can handle the full POSIX spec?
I do not see a "typical" user needing to access a path with say a network storage but multiple ../.. and hard and soft symlinks simultaneously.
I think "be liberal in what you accept" might need to be revisited for path parsing with some sort of OS-wide single-implementation as an optional feature.
What I mean is that for “honest” software, built-in to the OS or otherwise, the programmer finds a situation where they take some user-supplied input and concatenate that into a path, and call something like OS.read(). If they want to prevent the user from causing havoc, they now find themselves dealing with path validation in their software instead of calling OS.safeOpen(), which would be a reduced subset of allowed chars?
It is a lot of fixes and the Android Security Bulletins of June and Android 17 also had a lot of fixes [1], despite ASBs only containing high/critical vulnerabilities (other vulnerabilities are only fixed in major releases and QPRs, which most Android vendors respectively roll out late or never at all).
I think the story here is that vulnerability discovery has accelerated a lot with LLMs, but since are adversaries are doing the same, it is more important than ever to update quickly (and not let some Android vendors get away with their lazy update schedules).
Right - it was always dangerous but people who figured they weren’t important enough to be attacked might find out that LLMs have shifted that cost in the wrong direction.
And it's not actually that much information "about the security content". For example: "Impact: An app may be able to access sensitive user data. Description: An access issue was addressed with additional sandbox restrictions." This references CVE-2026-43819, which doesn't have any more information. Compare this with the nearly decade-old https://support.apple.com/en-gb/103680, and you see much more specific information about problems and their remedies (except in situations where Apple's action was to update a vendor component).
The vagueness could be intentional. There’s been a big issue with linux where proof of concept exploit code gets posted before the bug is announced because people reverse engineer it from the fix commits.
Apple has the advantage that they can keep everything secret for long enough for the patches to roll out. And realistically there is no reason the user needs to know the details of an exploit that was patched before it was ever used.
Collision counts are absurd. CVE-2026-43739 has roughly twenty credited researchers; CVE-2026-43816 has nearly as many. And ai attribution getting credit.
Apropos, anyone else saw "fast user switching" in Tahoe turn into "excruciatingly slow user switching which after a minute of switching without success rebooted the whole damn machine"?
15.7.8 is out today as well, with these security fixes: https://support.apple.com/en-us/128071.
For context, there have been issues with MacOS 26 which have led many people to defer upgrading until MacOS 27 is available, and MacOS 15 is the previous version.
> For context, there have been issues with MacOS 26 which have led many people to defer upgrading
For me the issue is liquid glass. Which I doubt is getting fixed any time soon
Also a liquid glass hater, but for what it's worth, 27 is supposed to make it a little bit better. I would prefer to go back to what I had before I had to upgrade into this horrible UI, but this is better than nothing.
https://www.cultofmac.com/news/liquid-glass-changes-ios-27-m...
I've been running the 27 beta and it fixes so much of the ugliness I hated in Tahoe. Lots of places had toolbars restored, solid areas that indicate where the window can be grabbed. The shitty icon spam in menus is back to sanity. And window corner radii are consistent and less bulbous.
macOS 17 is decisively faster on my MacBook Air M1 16 GB than macOS 26/27 is on my MacBook Pro M2 Max 32 GB. And I don’t mean just the animations themselves but lag/sluggishness/responsiveness in general.
There is no macOS 17.
Gosh I thought I had edited it to macOS 15, now it’s too late
Slower than 15 though
You can almost turn it off on the 27 beta, and 27 seems to perform a lot better too.
You can already turn off most glass effects in 26 via "System Settings => Accessibility => Display => Reduce Transparency".
I had that turned off years ago (for reasons I don't remember), and was wondering what all the fuzz was about when 26 came out because I didn't see much of a difference ;)
IMHO the actual important visual changes in the 27 beta is that rolls back the bizarre oversized corner radius in Finder windows, and they also got rid of the 'every menu item must have an icon' idea.
27 seems superior to 26 in almost every way, although I'm still on a fairly old beta.
I think that is just iOS no? Or does that work on MacOS too
It's the first thing I turn off on a new iPhone or iPad. Started around ten years ago in the release which had animated app icons. Induced nausea.
Yep. That’s why it’s in accessibility - some people can’t handle it. Other useful options are reduce motion, reduce transparency…
> this is better than nothing.
Having installed the beta, I think that's the best you can say about it.
Also on the beta. Agreed. Definite visual and UX improvement over 26, fixing the most egregious issues with 26 implementation of Liquid Glass, but maybe not quite as good as 25 overall.
> not quite as good as 25 overall
Nothing will ever be as good as 25.
Because macOS 25 does not exist. ;-)
Ahhh, my mistake. macOS 15 not to be confused with macOS 10.15
I hear it’s slower? Can you comment on that?
The 27 betas? No, they are generally faster than 26.
Definitely not slow.
I haven't noticed or heard that it's slower.
Nice, thanks for that.
As long as the UI improves and I can ignore all the AI stuff they're starting to push through, that's fine with me, though like many longtime macOS users, I'm not holding my breath for a bug-free experience.
> AI stuff they're starting to push through
iOS 26 anecdote:
A couple of weeks ago, I had a Baltimore Oriole (a cool-looking bird, not a baseball player) in my yard. They aren't rare, per se, but they are uncommon.
Took my iPhone out to snap a picture, and pressed the camera button. I hadn't used it, since upgrading to 26.
It takes the picture. It's there. I can see it, but it won't let me save it. Instead, it wants to tell me about the cool new voice-activated AI retouch feature. There was no way to save.
I probably could have figured it out, but I was so furious, I just nuked the picture.
> though like many longtime macOS users
For those of us older than just 10 years of using macOS, the older Apple OSes have instilled within us the desire to never install the X.0 release and wait until at least the X.1 release. The bug-free experience is a myth
> wait until at least the X.1 release
I think that applies to almost all software, not just Apple OSes. After all, Confucius said: "The only thing worse than old software is new software."
I found if you turn off transparency effects and turn on high contrast in the accessibility settings, it's actually a pretty nice looking UI. Still has some dumb quirks like inconsistent corner radiuses and text appearing under UI elements (because they're assumed to be transparent, but aren't). But all software is trash in 2026 so be thankful it isn't even worse, I guess.
Been doing that for years now already (performance boost). You can also do high contrast mode per app (if were talking about iOS). Must have for iMessage I think personally as it makes hard to read Green Bubbles a nice dark green that is very pleasant.
27 is a lot better. You can make the UI opaque again (they provide a transparency slider)
I bought a MBP mainly for the hardware, otherwise I'd have stuck with Linux. OS-wise, the jump from Monterey (which I'd last used) to Sequoia was smooth and still feels that way. I prefer not to notice the OS at all, something that I feel would be hard to do on Tahoe due to all that liquid glass and dumb transparency and animations.
Seriously, who the heck even asked for those?
macOS 27 does polish Liquid Glass and makes it look passable on macOS IMHO. It was very bad on 26. Comically bad.
That's the sad thing: Apple decided to leave all Intel macs on broken macOS 26. Very bad on Apple's part.
Fortunately, those can run Linux. I recently installed Arch on an Intel T2. The only issue is that it does not have a TPM module, so the LUKS password needs to be manually entered at boot.
What's the alternative to typing in a LUKS password?
I hate the round corners. It's already too much on 15, but way worse on 26. It looks like "Baby's first OS", designed by Fisher-Price.
They reverted this in 27
https://www.macrumors.com/2026/06/09/macos-golden-gate-liqui...
Love to see this, now if only they could revert the whole shimmering oil slick disaster that is "glass" on iOS (unlike macOS, even if I "reduce transparency", the oil slick effect still appears in many places, in addition to all the spacing issues).
Same. It's not the worst thing in the world, at least with Reduce Transparency enabled in the Accessibility settings, but I still don't feel any inclination to upgrade. My personal Mac Studio is macOS 15 and my work MacBook is on macOS 26, and I don't think there's a single thing that I find to be better on the work laptop than on my personal machine.
I might update to macOS 26 in September to be ready to update to macOS 27. Being two versions behind doesn't seem reasonable and I'd rather be on the "Tahoe but less shitty" version than Tahoe itself.
You can disable it in the accessibility preferences.
Same thing happens almost every release. I've stopped updating my Mac machine until I see something in the release notes I literally have to have in order to continue doing macOS/iOS builds, otherwise I'm staying on the version I've validated to work, and I know the existing bugs with.
As we used to say in the Windows world, wait for service pack 3.
A better strategy would probably be to stick with the previous *major* release, but, do install its ("minor") security updates...
> do install its ("minor") security updates
Yeah, I thought so too, but surprise surprise; some months ago one of the "minor" updates "broke" ("upgraded") something that made my CI/CD setup stop working, that's when I dropped the idea that Apple even do "minor" updates anymore.
>have led many people to defer upgrading until MacOS 27 is available
Then there's me, crying in MacBook Pro 2019 stuck on MacOS 15 because 27 won't be available for my machine.
The question now is, is 27 sufficient enough of an improvement over 15 to upgrade and tolerate Liquid Glass?
26.0 had a very annoying video jitter issue, but that was the first things that I noticed to be fixed in the next 26 release. Other than that, it worked just fine.
The bug that led to network connection issues after 49 days of uninterrupted uptime was a bit of a showstopper for me.
Are there no versions between MacOS 15 and 26???
They changed the numbering scheme, so... no, there aren't. Version numbers are now year-based, but previously they were not.
Last year they unified all their OS version numbers to just match upcoming year.
macOS went from 15 to 26
iOS went from 18 to 26
watchOS went from 11 to 26
and so on
No, 26 is the successor of 15. They changed the numbering to year-based.
Did they un-hardcode the corner radius? I mean were they able to? I mean not that that anyone at this point needs convincing how utterly disgusting incompetent they’re at software.
They reduced the radius back to older, smaller size and now all apps use a single radius, vs the weird 3 different radii in Tahoe. Looks better.
I was shocked to find out the inconsistent corner radii in Tahoe was an intentional design decision. I figured it's so obviously bad that it was just sloppy work. But one of the WWDC session videos bafflingly clarified that it depends on whether the app has a toolbar or not.
https://youtu.be/VqTn9NgiE1s?t=439
I can't imagine how the people who signed off on that were put in charge of design at Apple.
I am fairly convinced this was all designed only in graphic tools, and looked nice in static mockups, like those fake desktops with overlapping small windows widely spaced, etc. The design folks probably all had giant studio displays with plenty of room, etc.
Then the reality of "what about toolbars", "what about dark mode", "what about laptop screens", etc were all afterthoughts and resulted in bolt-on fixes like that.
Map the amount of fixes with "... improved bounds checking...", "...improved memory handling...", "...improved memory management..." into the amount of developer, QA and release management teams salaries per hour, versus other stuff they could be working on, and that gives an approximate value of how using specific languages maps into monetary loss, and why companies are starting to care nowadays, given computers are always exposed to the world network.
>>, and that gives an approximate value of how using specific languages maps into monetary loss, and why companies are starting to care nowadays, given computers are always exposed to the world network.
You need also factor development time and ease of finding developers willing to work in a specific language. There are other factors like readability of the code (very verbose languages are likely to be worse) and cost of maintenance - languages forcing a lot of abstractions are likely much worse.
> You need also factor development time and ease of finding developers willing to work in a specific language
This even more strongly favors Rust or Swift. Nobody is writing C or even Objective-C in 2026 as a growth language.
If you just meant apple-targeting developers, then yeah; you're right that those languages are in decline. But if you meant developers in general, I think you'd be surprised how many growth sectors are hiring C programmers. They're often not SaaS tech companies, but they are massive and many are growing. Hardware, industrial control systems, defense/aerospace ... there's a ton there, the spaces in which they hire just don't overlap a ton with the spaces frequented by hacker news. Also, a lot of them aren't US based companies.
I hope that changes over time, since I definitely agree that the downsides of C-family languages massively outweigh the downsides of competitor languages.
Which is why WG14 and WG21 caring about security would be quite relevant, but alas, priorities.
C could have gotten slices already in the 90's, the concept already existed in other languages, and even Dennis Ritchie made a fat pointer proposal into that sense.
The others, let see if anything related to profiles actually gets into C++29.
> very verbose languages are likely to be worse
Citation needed. I don't think there's a correlation there. Over-architected Java spaghetti is verbose and unmaintainable. Under-architected Perl code golf that metastisized is terse and unmaintainable.
> languages forcing a lot of abstractions are likely much worse
Citation needed. C++ has had some very high-level abstractions on top of a low-level runtime for awhile, and plenty of people have decided to use it and hire for it regardless. What counts as an "abstraction" or "forced abstraction" is a very very subjective topic.
Even C can be super abstracted, see early 1980-90's business software written in C, with nice stuff like Yourdon Structured Method, leading to over-architected C spaghetti with macros, a decade before Java came to be.
The problem is the lack of interest since Morris worm came to be, to provide better mechanisms in said languages, until governments and key big tech names decided it was time to change existing practices.
If anything, there's a strong argument to switch to seL4.
It was my vague understanding that by the time you implemented all the apis needed to run normal software on top of that, you either have enough apis that different tasks can still compromise each other, or you have shoved everything into a single task with very little isolation between normal user processes. In either case, it doesn't seem like you actually gained so much. What am I missing?
Indeed, however without some regulatory help it Will take its time for such kind of improvements across the industry.
“nah bro, all those other developers are just garbage, I am the one person that can write memory safe C”
Lots of "in collaboration with Claude and Anthropic Research" mentions, no mentions of other labs. I'd assume Apple already had access to whatever the most powerful model is at the various US-based labs, but perhaps not?
Those were voluntary disclosures by two Anthropic researchers and the security firm Calif. I know one more CVE on the list that was discovered using an AI agent and wasn't disclosed as such. I suspect there are many more.
Apple isn’t friends with OpenAI anymore
What happened?
https://www.cnbc.com/2026/07/10/apple-openai-lawsuit-trade-s...
The gist is, OpenAI hired a high ranking Apple employee who helped other Apple employees get hired by OpenAI and exfiltrate Apple trade secrets in the process.
Allegedly of course.
> high ranking Apple employee
Jony Ive basically works for Open AI (it's more complicated, but it's a good approximation), and has more or less rebuilt a designing team over there.
He's not the central person mentioned in Apple's accusations but that's arguably the central point that's triggering all of this.
Didn't he live Apple a very long time ago?
Yes, the high ranking employee at the center of the accusations is not Ive, but Tang Tan, former VP of product design
https://www.bbc.com/news/articles/cy8w379e091o
He left in 2019 and formed his own company, that did design work for Apple until 2022.
He "took" several Apple employees with him when he left and there's been a steady stream of Apple employees going to OpenAI.
Ive isn’t responsible for all of them obviously, but the articles about lawsuits says there are 400 former Apple employees at OpenAI.
Sounds like Apple needs to do a better job at being a place where employees want to stay.
Or, good that the craze for ever thinner laptops has been slowed?
A lot more than several. Sounded like it ended up being a large chunk of his team.
The accusations are pretty wild
https://news.ycombinator.com/item?id=48865294
> Lots of "in collaboration with Claude and Anthropic Research" mentions
I wouldn't say 4 is lots. The entire list is massive. I haven't counted myself, but someone claimed that macOS 26.6 has the all-time record with 155 CVEs.
Considering that in Feb 2026 (https://support.apple.com/en-us/126348) Claude wasn't mentioned even once, 4 sure sounds like "lots" compared to nothing :) But you're right, it's subjective ultimately.
Apple also hosts a copy of Claude internally in their servers.
Do they? As in Claude but on premises? Wonder if this is gonna be the solution that e.g. banks will require, exactly like they do now for cloud services (e.g. Azure on premises).
Banks are all about security theatre so probably not.
Pretty extreme solution… you can get Claude models from AWS Bedrock and Google Model Zoo. These are both very helpful for compliance and security, but do require you to have a cloud strategy.
Some data is so sensitive it likely has to stay on premises though.
That's why banks use Azure on Premises (not sure if other providers offer the same, but the investment bank I worked at did)
Yeah, albeit an increasingly second-rate experience, at least when it comes to Bedrock.
source?
edit: it seems asking for a source it frowned uppon in this site. And it seems there's no source.
I don't believe it's published anywhere, but it's common knowledge to Apple engineers. I can second the poster's assertion that they run Claude internally.
I know they use it, they host it too?
That would be a very Apple thing to do.
Yes, they host it on their own infrastructure.
Makes sense. I’m sure they have no interest in every Claude prompt going to Anthropic to keep eyes on what Apple is up to.
Yeah that's why I asked. It would be one thing to use, another to host.
And it seems nobody has a source so it's just humors as usual.
> That would be a very Apple thing to do.
That's something Steve Jobs would have done.
I don’t know about hosting it internally but a an Apple focused podcast I listen to (Accidental Tech Podcast) has mentioned numerous times in the last few months they’re using Claude heavily. And they know Apple insiders.
also “ Using GLM From Z.AI”
This may be a naive take, so if anyone has insight please feel free to share, but across Windows, Mac, and Linux OS's I see many cases of path parsing vulnerabilities resulting in sandbox escapes, code execution, or data access issues. When presenting the user with a file picker or command-line input, is it really needed that the software can handle the full POSIX spec?
I do not see a "typical" user needing to access a path with say a network storage but multiple ../.. and hard and soft symlinks simultaneously. I think "be liberal in what you accept" might need to be revisited for path parsing with some sort of OS-wide single-implementation as an optional feature.
> I do not see a "typical" user needing to access a path with say...
Typical users run software written by atypical users.
> some sort of OS-wide single-implementation
How do you propose handling migration? What if someone tries to expand an old archive file containing a now-forbidden path?
What I mean is that for “honest” software, built-in to the OS or otherwise, the programmer finds a situation where they take some user-supplied input and concatenate that into a path, and call something like OS.read(). If they want to prevent the user from causing havoc, they now find themselves dealing with path validation in their software instead of calling OS.safeOpen(), which would be a reduced subset of allowed chars?
If the OS is working properly, the havoc should just result in "permission denied."
If there's a path on the system that the user should not be able to read, that's the job of the OS to handle, not the individual applications.
How would you enforce a single implementation of path parsing?
Weird thing to see at number 3 on HN - is there some subtle context I am missing here?
Are we wink winking that it's a lot of fixes?
It is a lot of fixes and the Android Security Bulletins of June and Android 17 also had a lot of fixes [1], despite ASBs only containing high/critical vulnerabilities (other vulnerabilities are only fixed in major releases and QPRs, which most Android vendors respectively roll out late or never at all).
I think the story here is that vulnerability discovery has accelerated a lot with LLMs, but since are adversaries are doing the same, it is more important than ever to update quickly (and not let some Android vendors get away with their lazy update schedules).
[1] https://source.android.com/docs/security/bulletin/2026/2026-... https://source.android.com/docs/security/bulletin/android-17
So using newish phones that don't get updated anymore could be a lot more dangerous now than it was just a year ago.
Right - it was always dangerous but people who figured they weren’t important enough to be attacked might find out that LLMs have shifted that cost in the wrong direction.
That sucks.
And it's not actually that much information "about the security content". For example: "Impact: An app may be able to access sensitive user data. Description: An access issue was addressed with additional sandbox restrictions." This references CVE-2026-43819, which doesn't have any more information. Compare this with the nearly decade-old https://support.apple.com/en-gb/103680, and you see much more specific information about problems and their remedies (except in situations where Apple's action was to update a vendor component).
The vagueness could be intentional. There’s been a big issue with linux where proof of concept exploit code gets posted before the bug is announced because people reverse engineer it from the fix commits.
Apple has the advantage that they can keep everything secret for long enough for the patches to roll out. And realistically there is no reason the user needs to know the details of an exploit that was patched before it was ever used.
Relevant context might be for example that there are 4 mentions each of Claude by Anthropic and XGPT by ThreatBook, both based on LLMs.
AI attribution might be one reason people are particularly curious.
I missed that, thanks for pointing out
I think it's because it's the first big batch of fixes found at Apple by Mythos.
Is this speculation? Where does it say Mythos was responsible for any of this?
It is speculation, which is what I think the upvote count reflects.
Yes, it's really quite a lot https://9to5mac.com/2026/07/27/ios-26-6-fixes-over-75-securi...
I think that's it?
Collision counts are absurd. CVE-2026-43739 has roughly twenty credited researchers; CVE-2026-43816 has nearly as many. And ai attribution getting credit.
One CVE even lists the same person twice!
CVE-2026-64691: Ruslan Dautov, Ruslan Dautov
> One CVE even lists the same person twice!
Not necessarily. Could be two persons sharing that name. See https://revstat.ine.pt/index.php/REVSTAT/article/view/382
How did you unearth such a remarkable find of 2 different faculty individuals named "Philipp Otto" collaborating on the same paper :) ?
One of them lists an anonymous person!
CVE-2026-43744: Mathis Mansière, an anonymous researcher
It reads as if "an anonymous researcher" is describing Mathis Mansière, which is quite humorous.
This reminds me of my favorite segment of the TV show Curb Your Enthusiasm: https://youtu.be/JqrJ4wGid4Y
Spell checker fixed a typo, it was originally "an Anonymous researcher" /s
Genius, that made my day!
It's Ted Danson.
Apropos, anyone else saw "fast user switching" in Tahoe turn into "excruciatingly slow user switching which after a minute of switching without success rebooted the whole damn machine"?
It would be so nice to see how many zero-days are going away for bad players right now.