They made C memory safe? This is a big thing to gloss over in a single paragraph. Does anyone have extra details on this?
> On devices with iOS 14 and iPadOS 14 or later, Apple modified the C compiler toolchain used to build the iBoot bootloader to improve its security. The modified toolchain implements code designed to prevent memory- and type-safety issues that are typically encountered in C programs. For example, it helps prevent most vulnerabilities in the
following classes:
> • Buffer overflows, by ensuring that all pointers carry bounds information that’s verified
when accessing memory
> • Heap exploitation, by separating heap data from its metadata and accurately detecting error conditions such as double free errors
> • Type confusion, by ensuring that all pointers carry runtime type information that’s verified during pointer cast operations
> • Type confusion caused by use after free errors, by segregating all dynamic memory allocations by static type
Sort of. From my understanding they’ve been heavily using clang with fbounds checks to insert checks into functions. I think there was work done to try to insert them into existing code as well. They memory tagging in new processors help avoid overflow exploitation. Maybe someone can jump in and add more details
Apple's commitment to privacy and security is really cool to see. It's also an amazing strategic play that they are uniquely in the position to take advantage of. Google and Meta can't commit to privacy because they need to show you ads, whereas Apple feels more like a hardware company to me.
1. Google defaults to encrypted backups of messages, as well as e2e encryption of messages.
2. Apple defaults only to e2ee of messages, leaving a massive backdoor.
3. Closing that backdoor is possible for the consumer, by enabling ADP (advanced data protection) on your device. However, this makes no difference, since 99.9% of the people you communicate will not close the backdoor. Thus, the only way to live is to assume that all the messages you send via iMessage will always be accessible to Apple, no matter what you do.
It's not like overall I think Google is better for privacy than Apple, but this choice by Apple is really at odds with their supposed emphasis on privacy.
I still like to encourage people to watch all of https://www.youtube.com/watch?v=BLGFriOKz6U&t=1993s for the details (from Apple’s head of Security Engineering and Architecture) about how iCloud is protected by HSMs, rate limits, etc. but especially the timelinked section. :)
What matters is that the parent comment said “Apple is an ad company now,” as if that negated all the privacy and security stuff they do.
Making some cash on ads doesn’t have to rely on targeted tracking. That only matters if ads are an existential part of your business, and without huge ad revenue growth, your company is dead.
I mean if you don’t care about details that’s fine I guess. Let’s call any company that sells and/or buys any amount of ads an "ad company" and put them all into one bucket and judge. That’s super valuable.
I claim bs at this whole apple privacy thing, nothing but propaganda.
Two years ago I was locked out of my MacBook pro.
Then I just booted in some recovery mode and just..reset the password!?
Sure macos logged me off from (most) apps and website, but every single file was there unencrypted!
I swear people that keep boasting that whole apple privacy thing have absolutely no clue what they are talking about, nothing short of tech illiterate charlatans. But God the propaganda works.
Apple has since confirmed in a statement provided to Ars that the US federal government “prohibited” the company “from sharing any information,” but now that Wyden has outed the feds, Apple has updated its transparency reporting and will “detail these kinds of requests” in a separate section on push notifications in its next report.
That people fall for this corporate BS while Tim Cook is giving gold bars to Trump and dining and dancing with him When people are being murdered on the streets by ice is just amazing to me.
Yeah Americans voted for Trump. But that shouldn't prevent CEOs to show a spine. Tim Cook is no different from all the others, therefore Apple doesn't deserve any less contempt from us.
Funny that you think that people have free will under this zombie social media mind controlled Internet world we’re living in.
Besides Trump‘s approval ratings are worse than ever so I don’t think people really got what they wanted, they got who they voted for not what they voted for.
> Tim Cook was (supposedly) principled. I guess it's hard to pretend that you care about privacy or human rights while eating dinner next to bin Salman.
I guess if you thought he had principles then yeah that could be disappointing. Personally I've never tried to moralize corporations though, I just assume the only principle that every company and CEO operates by is whatever increases the stock price.
You know this is just marketing right? Apple gives zero fucks about security. They just use it to lock competitors out of their gardens and preach a holier-than-thou attitude about it.
All while slowly stuffing (more?) ads into their software.
In a lot of ways Apple is as aligned to data privacy the same way other "platforms" are: to gatekeep the user data behind their ad service. It's better than selling your data, maybe, but you're still being tracked and monitored.
The worst part is since Apple is technically not a 3rd party, many of the rules don’t apply to them even though they bring the same harm to the users. Did you notice the new “creative suite” has analytics with identities linked to your Apple account turned on by defend? Free Pages/Numbers is not so free anymore.
We know now that it was all marketing talk. Apple didn’t like Meta so they spun a bunch of obstacles. Apple has and would use your data for ads, models and anything that keeps the shareholders happy. And we don’t know the half of the story where as a US corp, they’re technically obliged to share data from the not-E2EE iCloud syncs of every iPhone.
You can request a downloadable a copy of any/all of the data that Apple has associated with your account at https://privacy.apple.com.
This apparently includes retrieving all photos from iCloud in chunks of specified size, which seems an infinitely better option than attempting to download them through the iCloud web interface which caps downloads to 1000 photos at a time at less than impressive download speeds.
But all the software is closed source, and there is little to no opportunity to verify all these security claims. You don't have the encryption keys, so effectively the data is not under your control.
If you want to see security done well (or at least better), see the GrapheneOS project.
GrapheneOS also doesn't give you the encryption keys. If you run the official version, there is no way for you to extract the data from your device at all beyond what app developers will let you access. This means that you do not own the data on your device. The backups are even less effective than Apple's, although they say they will work on it.
The developers also appear to believe that the apps have a right to inspect the trustworthiness of the user's device, by offering to support apps that would trust their keys [1], locking out users who maintain their freedom by building their own forks.
It's disheartening that a lot of security-minded people seem to be fixated on the "AOSP security model", without realizing or ignoring the fact that a lot of that security is aimed at protecting the apps from the users, not the other way around. App sandboxing is great, but I should still be able to see the app data, even if via an inconvenient method such as the adb shell.
> The developers also appear to believe that the apps have a right to inspect the trustworthiness of the user's device, by offering to support apps that would trust their keys [1], locking out users who maintain their freedom by building their own forks.
That is not a bad thing. The alternative is not having apps that do these checks available on the platform at all. It’s ridiculous that someone should expect that every fork of it should have that capability (because the average developer is not going to accept the keys of someone’s one off fork).
If there’s anyone to blame, it should be the app developers choosing to do that (benefits of attestation aside).
Attestation is also a security feature, which is one of the points of GOS. People are free to use any other distribution of Android if they take issue with it.
Obviously I could be wrong here, this is just the general sentiment that I get from reading GOS documentation and its developer’s comments.
I don't actually disagree with this. The auditor is a perfectly valid use of it. It's good to be able to verify cryptographically your device is running what it's supposed to.
The problem is when it transcends ownership boundaries and becomes a mechanism to exert control over things someone doesn't own, like your bank or government controlling your phone. It is one of the biggest threats to ownership worldwide.
Note also that getting "trusted" comes at the cost of other security features, such as spoofing your location securely to apps:
You were not going to be able to use those apps anyways, so what does it matter to you? I, and I suspect many, agree with the purpose of attestation. The problems around it are strictly around establishing good ways to teach apps who they should trust, not around attestation itself. By putting your head in the sand, you'll never improve the situation.
Ah, the apps^Wgovernment (look at that page, most of it is government IDs) should be able to discriminate against me for daring to assert control over my own device. And GrapheneOS is saying:
Hey government! We pinky promise to oppress the user just the same, but even more securely and competently than Google/Samsung!
> what does it matter to you
It shows that the developers maybe don't fully have your best interests at heart?
It sucks that Apple decided to monitize iPhone the way they have, by controlling the owners ability to install software of their choosing. Ignoring the arguments one could make about this making it "more secure" it's clearly disrespectful to the power user that doesn't want to beg Apple's permission to use their computer. I'll grant them their security claims are sound, but it's hard to take them serious regarding privacy arguments.
Our choices are either (A) an OS monitized by tracking user interaction and activity, or (B) monitized by owning the basic act of installing software on the device, both of these options suck and I struggle to give up the more open option for one that might be more secure.
Ignoring the arguments one could make about this making it "more secure" it's clearly disrespectful to the power user that doesn't want to beg Apple's permission to use their computer. I'll grant them their security claims are sound,
I wouldn't say they are sound. First, macOS provides the freedom to install your own applications (ok, they need to be signed and notarized if the quarantine attribute is set) and it's not the case that the Mac has mass malware infestations. Second, the App Store is full of scams, so App Store - safe, external - unsafe is a false dichotomy.
Apple uses these arguments, but of course the real reason is that they want to continue to keep 30% of every transaction made on an iPhone or iPad. This is why they have responded to the DMA with a lot of malicious compliance that makes it nearly impossible to run an alt-store financially.
(Despite my qualms about not being able to install apps outside the app store, I do think they are doing a lot of good work of making the platform more secure.)
Given that A19 + M5 processors with MIE (EMTE) were only recently introduced, I wonder how extensively MacOS/iOS make use of the hardware features. Is it something that's going to take several years to see the benefit, or does MIE provide thorough protection today?
Apple’s implementation of MTE is relatively limited in scope compared to GrapheneOS (and even stock Android with advanced security enabled) as it’s hardware intensive and degrades performance. I imagine once things get fast enough we could see synchronous MTE enabled everywhere.
It is curious at the moment though that enabling something like Lockdown Mode doesn’t force MTE everywhere, which imo it should. I think the people who are willing to accept the compromises of enabling that would likely also be willing to tolerate the app crashes, worse performance etc that would come with globally enabled MTE.
I think all of the kernel allocators and most (?) system processes in iOS 26 have MIE enabled, as does libpas (the WebKit allocator), so it’s already doing quite a lot.
The ones I remember most affecting performance were zeroing allocated memory and the Spectre/Meltdown fix. Also, the first launch of a new app is slow in order to check the signature. Whole disk encryption is pretty fast today, but probably is a bit slower than unencrypted. The original FileVault using disk images was even slower.
It's not really possible to make a direct comparison, given that a big chunk of the features are baked into the silicon, or are architecture-level choices.
It’s technically possible, but it would be difficult and likely require breaching an NDA. A bit pedantic, perhaps, but it’s out there.
Apple makes available on a highly controlled basis iPhones which permit the user to disable “virtually all” of the security features. They’re available only to vetted security researchers who apply for one, often under some kind of sponsorship, and they’re designed to obviously announce what they are. For example they are engraved on the sides with “Confidential and Proprietary. Property of Apple”.
They’re loaned, not sold or given, remain Apple’s property, and are provided on a 12-month (optionally renewable) basis. You have to apply and be selected by Apple to receive one, and you have to agree to some (understandable but) onerous requirements laid out in an legal agreement.
I expect that if you were to interrogate these iPhones they would report that the CPU fuse state isn’t “Production” like the models that are sold.
They refer to these iPhones as Security Research Devices, or SRDs.
Then they turn around and upload your iMessages to their own servers in a form that they can read, breaking their own E2EE. Google Messages fixed this issue a long time ago. Why hasn't Apple? https://james.darpinian.com/blog/apple-imessage-encryption
What is "Google Messages"? I can't count the number of articles people have written over time about how many first-party messaging apps Google themselves have put out (and then put down), not to mention what messaging apps get shoveled on by third-party android integrators.
> the main reason a message wouldn't be properly end-to-end encrypted in Google's Messages app is when communicating with an iPhone user, because Apple has dragged their feet on implementing RCS features in iMessage
(or with any other android user who isn't using a first-party device / isn't using this one app)
> [...] Android's equivalent cloud backup service has been properly end-to-end encrypted by default for many years. Meaning that you don't need to convince the whole world to turn on an optional feature before your backups can be fully protected.
You make it out to seem that it's impossible for Google to read your cloud backups, but the article you link to [0] earlier in your post says that "this passcode-protected key material is encrypted to a Titan security chip on our datacenter floor" (emphasis added). So they have your encrypted cloud backup, and the only way to get the key material to decrypt it is to get it from an HSM in their datacenter, every part of which and the access to which they control... sounds like it's not really any better than Apple, from what I'm reading here. Granted, that article is from 2018 and I certainly have not been keeping up on android things.
HSMs are designed to protect encryption keys from everyone including the manufacturer. Signal trusts them for their encryption features. It's the best security possible for E2EE backups with passcode recovery, and Apple does it too for the subset of data that they do real E2EE backups on, like Keychain passwords. Characterizing using an HSM to implement E2EE securely as "not any better than" just giving up on E2EE for messages backups is ridiculous.
The HSMs that Signal and Apple are using are on-device though. Yes you still have to trust Signal / Apple to not exfil your key matter once decrypted by the HSM, but I submit that that is materially better than having the HSMs be hosted in a datacenter.
You can enable Advanced Data Protection to address that issue with iMessages.
Giving users an option between both paths is usually best. Most users care a lot more that they can’t restore a usable backup of their messages than they do that their messages are unreadable by the company storing them.
I used to work at a company where our products were built around encryption. Users here on HN are not the norm. You can’t trust that most users will save recovery codes, encryption seed phrases, etc in a manner that will be both available and usable when they need them, and then they tend to care a lot less about the privacy properties that provides and a lot more that they no longer have their messages with {deceased spouse, best friend, business partner, etc}.
> Apple can still read any message you exchange with practically anyone through their iCloud backups, since they are overwhelmingly likely to have backups enabled and overwhelmingly unlikely to have proactively enabled the non-default "Advanced Data Protection" feature.
> They could have implemented iMessage to not backup messages from people who enabled ADP, but they didn't. They won't even inform you when your conversation partner has uploaded your messages to Apple's servers in a form that Apple can read.
> Android's equivalent cloud backup service has been properly end-to-end encrypted by default for many years. Meaning that you don't need to convince the whole world to turn on an optional feature before your backups can be fully protected.
> Apple's stated reason for not enabling end-to-end encryption on iCloud backups by default is that it would cause data loss when users lose their devices. But Google's implementation avoids this problem. Furthermore, Apple does do end-to-end encryption by default on other critical information that would be painful to lose, such as your account passwords stored in Keychain. So that excuse doesn't seem to hold water.
This is your blog post, so I'll ask you a question. What are you trying to state in Belief #1? The message is unclear to me with how it's worded:
> In this table, in the "iCloud Backup (including device and Messages backup)" row, under "Standard data protection",
> the "Encryption" column reads "In transit & on server". Yes, this means that Apple can read all of your messages
> out of your iCloud backups.
In addition to the things you mentioned, there's certainly a possibility of Apple attaching a virtual "shadow" device to someone's Apple ID with something like a hide_from_customer type flag, so it would be invisible to the customer.
This shadow device would have it's own keys to read messages sent to your iCloud account. To my knowledge, there's nothing in the security model to prevent this.
This shadow device would have it's own keys to read messages sent to your iCloud account. To my knowledge, there's nothing in the security model to prevent this.
Matthew Green has some great posts about iMessage security. This one describes the key lookup issue:
Or Apple can also push an update, which you can't refuse, that upon first message to iCloud just uploads your private key. It's a bit foolish to count on encryption implemented by the adversary you're trying to hide from. Of course, this will most likely only affect individuals targeted by state-level actors.
The table has two categorizations: "In transit & on server" and "End-to-end". The former, which covers iCloud backups in the default configuration, is explicitly NOT end-to-end, meaning there are moments in time during processing where the data is not encrypted.
However, iCloud backups actually are listed as "End-to-end" if you turn on the new Advanced Data Protection feature.
Pegasus isn't magic. It exploits security vulnerabilities just like everything else. Mitigating and fixing those vulnerabilities is a major part of this document.
Why? The obvious conclusion is that Apple is doing everything in its power to make the answer “no.”
You might as well enumerate all the viruses ever made on Windows, point to them, and then ask why Microsoft isn’t proving they’ve shut them all down yet in their documents.
That analogy misses the asymmetry in claims and power.
Microsoft does not sell Windows as a sealed, uncompromisable appliance. It assumes a hostile environment, acknowledges malware exists, and provides users and third parties with inspection, detection, and remediation tools. Compromise is part of the model.
Apple’s model is the opposite. iOS is explicitly marketed as secure because it forbids inspection, sideloading, and user control. The promise is not “we reduce risk”, it’s “this class of risk is structurally eliminated”. That makes omissions meaningful.
So when a document titled Apple Platform Security avoids acknowledging Pegasus-class attacks at all, it isn’t comparable to Microsoft not listing every Windows virus. These are not hypothetical threats. They are documented, deployed, and explicitly designed to bypass the very mechanisms Apple presents as definitive.
If Apple believes this class of attack is no longer viable, that’s worth stating. If it remains viable, that also matters, because users have no independent way to assess compromise. A vague notification that Apple “suspects” something, with no tooling or verification path, is not equivalent to a transparent security model.
The issue is not that Apple failed to enumerate exploits. It’s that the platform’s credibility rests on an absolute security narrative, while quietly excluding the one threat model that contradicts it. In other words Apple's model is good old security by obscurity.
Apple did create a boolean for that. They call it lockdown mode.
> Lockdown Mode is an optional, extreme protection that’s designed for the very few individuals who, because of who they are or what they do, might be personally targeted by some of the most sophisticated digital threats. Most people are never targeted by attacks of this nature.
When Lockdown Mode is enabled, your device won’t function like it typically does. To reduce the attack surface that potentially could be exploited by highly targeted mercenary spyware, certain apps, websites, and features are strictly limited for security and some experiences might not be available at all.
If Pegasus can break the iOS security model, there’s no reason to think it politely respects Lockdown Mode.
It’s basically an admission the model failed, with features turned off so users feel like they’re doing something about it.
Lockdown mode works by reducing the surface area of possible exploits. I don't think there's any failures here. Apple puts a lot of effort into resolving web-based exploits, but they can also prevent entire classes of exploits by just blocking you from opening any URL in iMessage. It's safer, but most users wouldn't accept that trade-off.
They made C memory safe? This is a big thing to gloss over in a single paragraph. Does anyone have extra details on this?
> On devices with iOS 14 and iPadOS 14 or later, Apple modified the C compiler toolchain used to build the iBoot bootloader to improve its security. The modified toolchain implements code designed to prevent memory- and type-safety issues that are typically encountered in C programs. For example, it helps prevent most vulnerabilities in the following classes:
> • Buffer overflows, by ensuring that all pointers carry bounds information that’s verified when accessing memory
> • Heap exploitation, by separating heap data from its metadata and accurately detecting error conditions such as double free errors
> • Type confusion, by ensuring that all pointers carry runtime type information that’s verified during pointer cast operations
> • Type confusion caused by use after free errors, by segregating all dynamic memory allocations by static type
Sort of. From my understanding they’ve been heavily using clang with fbounds checks to insert checks into functions. I think there was work done to try to insert them into existing code as well. They memory tagging in new processors help avoid overflow exploitation. Maybe someone can jump in and add more details
Apple's commitment to privacy and security is really cool to see. It's also an amazing strategic play that they are uniquely in the position to take advantage of. Google and Meta can't commit to privacy because they need to show you ads, whereas Apple feels more like a hardware company to me.
modeless linked to this article earlier today:
https://james.darpinian.com/blog/apple-imessage-encryption/
My current understanding of the facts:
1. Google defaults to encrypted backups of messages, as well as e2e encryption of messages.
2. Apple defaults only to e2ee of messages, leaving a massive backdoor.
3. Closing that backdoor is possible for the consumer, by enabling ADP (advanced data protection) on your device. However, this makes no difference, since 99.9% of the people you communicate will not close the backdoor. Thus, the only way to live is to assume that all the messages you send via iMessage will always be accessible to Apple, no matter what you do.
It's not like overall I think Google is better for privacy than Apple, but this choice by Apple is really at odds with their supposed emphasis on privacy.
I still like to encourage people to watch all of https://www.youtube.com/watch?v=BLGFriOKz6U&t=1993s for the details (from Apple’s head of Security Engineering and Architecture) about how iCloud is protected by HSMs, rate limits, etc. but especially the timelinked section. :)
I still recommend Mr. Fart's Favorite Colors as a refutation, describing why all of these precautions cannot protect you in a real-world security model: https://medium.com/@blakeross/mr-fart-s-favorite-colors-3177...
Krstić: “Here’s how we reduce the chance that even Apple can access or alter X, and here’s how we can make that credible.”
Ross: “Even if you make X cryptographically airtight, the real fight becomes political/physical coercion: ‘ship this or else.’”
Those can both be true at the same time.
It's all tempered by them ultimately controlling what you can put on your phone though.
As was demonstrated in LA, it's starting to have significant civil rights consequences.
Apple is an ad company now though
Apple sells some ads yes. But it’s a tiny fraction of their revenue.
Would Google or Meta go bankrupt if they stopped selling ads? Yes. Apple wouldn’t.
What does whether they’d go bankrupt or not have to do with whether they’re an ad company?
They sell third party ads: companies unaffiliated with Apple pay Apple to advertise on Apple platforms.
They’re an ad company. Just because it’s currently a small slice of their total revenue doesn’t make it untrue.
I guess it’s also a financial company, since they have a branded credit card?
What matters is that the parent comment said “Apple is an ad company now,” as if that negated all the privacy and security stuff they do.
Making some cash on ads doesn’t have to rely on targeted tracking. That only matters if ads are an existential part of your business, and without huge ad revenue growth, your company is dead.
I mean if you don’t care about details that’s fine I guess. Let’s call any company that sells and/or buys any amount of ads an "ad company" and put them all into one bucket and judge. That’s super valuable.
Apple would go bankrupt without US protectionist policy propping up their service revenue.
That's pretty bad. Maybe not "reliant on ad monopoly" bad, but pretty close.
In their revenue report this week out of $140B, services made up 30B. 140B-30B = 110B. Thats pretty far from bankruptcy.
I claim bs at this whole apple privacy thing, nothing but propaganda.
Two years ago I was locked out of my MacBook pro.
Then I just booted in some recovery mode and just..reset the password!?
Sure macos logged me off from (most) apps and website, but every single file was there unencrypted!
I swear people that keep boasting that whole apple privacy thing have absolutely no clue what they are talking about, nothing short of tech illiterate charlatans. But God the propaganda works.
And don't start me on iMessage.
You chose not to enable FileVault during setup. Probably because you were worried about being locked out and wanted an easy way to reset the password.
Would you prefer that Apple did not give you the option to disable the security feature you disabled during setup?
Ain't nobody paying attention, in any case it's still propaganda.
You know what's even cooler? Apple's commitment to hiding US federally-mandated backdoors for dragnet surveillance: https://arstechnica.com/tech-policy/2023/12/apple-admits-to-...
Apple has ads. See the App Store, Apple Maps is also planning to roll out advertising.
That people fall for this corporate BS while Tim Cook is giving gold bars to Trump and dining and dancing with him When people are being murdered on the streets by ice is just amazing to me.
Well that’s what Americans voted for. So I don’t think anyone cares that every CEO (definitely not just Tim Cook) is schmoozing with Trump.
Yeah Americans voted for Trump. But that shouldn't prevent CEOs to show a spine. Tim Cook is no different from all the others, therefore Apple doesn't deserve any less contempt from us.
Funny that you think that people have free will under this zombie social media mind controlled Internet world we’re living in.
Besides Trump‘s approval ratings are worse than ever so I don’t think people really got what they wanted, they got who they voted for not what they voted for.
The Twinkie defense is alive and well, I see.
> Well that’s what Americans voted for.
Americans are not one person.
> So I don’t think anyone cares
Clearly they do.
> every CEO (definitely not just Tim Cook) is schmoozing with Trump.
Tim Cook was (supposedly) principled. I guess it's hard to pretend that you care about privacy or human rights while eating dinner next to bin Salman.
> Tim Cook was (supposedly) principled. I guess it's hard to pretend that you care about privacy or human rights while eating dinner next to bin Salman.
I guess if you thought he had principles then yeah that could be disappointing. Personally I've never tried to moralize corporations though, I just assume the only principle that every company and CEO operates by is whatever increases the stock price.
You know this is just marketing right? Apple gives zero fucks about security. They just use it to lock competitors out of their gardens and preach a holier-than-thou attitude about it.
All while slowly stuffing (more?) ads into their software.
In a lot of ways Apple is as aligned to data privacy the same way other "platforms" are: to gatekeep the user data behind their ad service. It's better than selling your data, maybe, but you're still being tracked and monitored.
The worst part is since Apple is technically not a 3rd party, many of the rules don’t apply to them even though they bring the same harm to the users. Did you notice the new “creative suite” has analytics with identities linked to your Apple account turned on by defend? Free Pages/Numbers is not so free anymore.
> Apple gives zero fucks about security.
Hyperbole doesn’t help your point. They definitely care about security, their profits depend on it.
> Apple's commitment to privacy
We know now that it was all marketing talk. Apple didn’t like Meta so they spun a bunch of obstacles. Apple has and would use your data for ads, models and anything that keeps the shareholders happy. And we don’t know the half of the story where as a US corp, they’re technically obliged to share data from the not-E2EE iCloud syncs of every iPhone.
You can request a downloadable a copy of any/all of the data that Apple has associated with your account at https://privacy.apple.com.
This apparently includes retrieving all photos from iCloud in chunks of specified size, which seems an infinitely better option than attempting to download them through the iCloud web interface which caps downloads to 1000 photos at a time at less than impressive download speeds.
Web version: https://support.apple.com/guide/security/welcome/web
That’s Dec 2024
But all the software is closed source, and there is little to no opportunity to verify all these security claims. You don't have the encryption keys, so effectively the data is not under your control.
If you want to see security done well (or at least better), see the GrapheneOS project.
GrapheneOS also doesn't give you the encryption keys. If you run the official version, there is no way for you to extract the data from your device at all beyond what app developers will let you access. This means that you do not own the data on your device. The backups are even less effective than Apple's, although they say they will work on it.
The developers also appear to believe that the apps have a right to inspect the trustworthiness of the user's device, by offering to support apps that would trust their keys [1], locking out users who maintain their freedom by building their own forks.
It's disheartening that a lot of security-minded people seem to be fixated on the "AOSP security model", without realizing or ignoring the fact that a lot of that security is aimed at protecting the apps from the users, not the other way around. App sandboxing is great, but I should still be able to see the app data, even if via an inconvenient method such as the adb shell.
1. https://grapheneos.org/articles/attestation-compatibility-gu...
> The developers also appear to believe that the apps have a right to inspect the trustworthiness of the user's device, by offering to support apps that would trust their keys [1], locking out users who maintain their freedom by building their own forks.
That is not a bad thing. The alternative is not having apps that do these checks available on the platform at all. It’s ridiculous that someone should expect that every fork of it should have that capability (because the average developer is not going to accept the keys of someone’s one off fork).
If there’s anyone to blame, it should be the app developers choosing to do that (benefits of attestation aside).
Attestation is also a security feature, which is one of the points of GOS. People are free to use any other distribution of Android if they take issue with it.
Obviously I could be wrong here, this is just the general sentiment that I get from reading GOS documentation and its developer’s comments.
> Attestation is also a security feature
I don't actually disagree with this. The auditor is a perfectly valid use of it. It's good to be able to verify cryptographically your device is running what it's supposed to.
The problem is when it transcends ownership boundaries and becomes a mechanism to exert control over things someone doesn't own, like your bank or government controlling your phone. It is one of the biggest threats to ownership worldwide.
Note also that getting "trusted" comes at the cost of other security features, such as spoofing your location securely to apps:
https://news.ycombinator.com/item?id=44685283
For some reason they don't release userdebug versions which was a dealbreaker for me.. (the device should be secure, but not against me)
But if you wish to build it from source, it could probably be a good option.
You can re-sign it using https://github.com/chenxiaolong/avbroot
I don't currently have any root on the phone, but I reserve the right to add it or run the userdebug build at a later date
You were not going to be able to use those apps anyways, so what does it matter to you? I, and I suspect many, agree with the purpose of attestation. The problems around it are strictly around establishing good ways to teach apps who they should trust, not around attestation itself. By putting your head in the sand, you'll never improve the situation.
> teach apps who they should trust
Ah, the apps^Wgovernment (look at that page, most of it is government IDs) should be able to discriminate against me for daring to assert control over my own device. And GrapheneOS is saying:
Hey government! We pinky promise to oppress the user just the same, but even more securely and competently than Google/Samsung!
> what does it matter to you
It shows that the developers maybe don't fully have your best interests at heart?
Yes, how can we verify this? Who says three-letter agencies have no access?
Glad there's still at least one tech company that cares about personal security / opsec.
It sucks that Apple decided to monitize iPhone the way they have, by controlling the owners ability to install software of their choosing. Ignoring the arguments one could make about this making it "more secure" it's clearly disrespectful to the power user that doesn't want to beg Apple's permission to use their computer. I'll grant them their security claims are sound, but it's hard to take them serious regarding privacy arguments.
Our choices are either (A) an OS monitized by tracking user interaction and activity, or (B) monitized by owning the basic act of installing software on the device, both of these options suck and I struggle to give up the more open option for one that might be more secure.
Ignoring the arguments one could make about this making it "more secure" it's clearly disrespectful to the power user that doesn't want to beg Apple's permission to use their computer. I'll grant them their security claims are sound,
I wouldn't say they are sound. First, macOS provides the freedom to install your own applications (ok, they need to be signed and notarized if the quarantine attribute is set) and it's not the case that the Mac has mass malware infestations. Second, the App Store is full of scams, so App Store - safe, external - unsafe is a false dichotomy.
Apple uses these arguments, but of course the real reason is that they want to continue to keep 30% of every transaction made on an iPhone or iPad. This is why they have responded to the DMA with a lot of malicious compliance that makes it nearly impossible to run an alt-store financially.
(Despite my qualms about not being able to install apps outside the app store, I do think they are doing a lot of good work of making the platform more secure.)
Given that A19 + M5 processors with MIE (EMTE) were only recently introduced, I wonder how extensively MacOS/iOS make use of the hardware features. Is it something that's going to take several years to see the benefit, or does MIE provide thorough protection today?
I was just watching a video on this yesterday: https://www.youtube.com/watch?v=5McB6-2r-ds
Apple’s implementation of MTE is relatively limited in scope compared to GrapheneOS (and even stock Android with advanced security enabled) as it’s hardware intensive and degrades performance. I imagine once things get fast enough we could see synchronous MTE enabled everywhere.
It is curious at the moment though that enabling something like Lockdown Mode doesn’t force MTE everywhere, which imo it should. I think the people who are willing to accept the compromises of enabling that would likely also be willing to tolerate the app crashes, worse performance etc that would come with globally enabled MTE.
I think all of the kernel allocators and most (?) system processes in iOS 26 have MIE enabled, as does libpas (the WebKit allocator), so it’s already doing quite a lot.
Sometime I wonder how much overhead all these security features take in terms of performance.
I would really like to see a benchmark with and without security measures.
The ones I remember most affecting performance were zeroing allocated memory and the Spectre/Meltdown fix. Also, the first launch of a new app is slow in order to check the signature. Whole disk encryption is pretty fast today, but probably is a bit slower than unencrypted. The original FileVault using disk images was even slower.
It's not really possible to make a direct comparison, given that a big chunk of the features are baked into the silicon, or are architecture-level choices.
It’s technically possible, but it would be difficult and likely require breaching an NDA. A bit pedantic, perhaps, but it’s out there.
Apple makes available on a highly controlled basis iPhones which permit the user to disable “virtually all” of the security features. They’re available only to vetted security researchers who apply for one, often under some kind of sponsorship, and they’re designed to obviously announce what they are. For example they are engraved on the sides with “Confidential and Proprietary. Property of Apple”.
They’re loaned, not sold or given, remain Apple’s property, and are provided on a 12-month (optionally renewable) basis. You have to apply and be selected by Apple to receive one, and you have to agree to some (understandable but) onerous requirements laid out in an legal agreement.
I expect that if you were to interrogate these iPhones they would report that the CPU fuse state isn’t “Production” like the models that are sold.
They refer to these iPhones as Security Research Devices, or SRDs.
These devices still have all the security features.
Then they turn around and upload your iMessages to their own servers in a form that they can read, breaking their own E2EE. Google Messages fixed this issue a long time ago. Why hasn't Apple? https://james.darpinian.com/blog/apple-imessage-encryption
What is "Google Messages"? I can't count the number of articles people have written over time about how many first-party messaging apps Google themselves have put out (and then put down), not to mention what messaging apps get shoveled on by third-party android integrators.
> the main reason a message wouldn't be properly end-to-end encrypted in Google's Messages app is when communicating with an iPhone user, because Apple has dragged their feet on implementing RCS features in iMessage
(or with any other android user who isn't using a first-party device / isn't using this one app)
> [...] Android's equivalent cloud backup service has been properly end-to-end encrypted by default for many years. Meaning that you don't need to convince the whole world to turn on an optional feature before your backups can be fully protected.
You make it out to seem that it's impossible for Google to read your cloud backups, but the article you link to [0] earlier in your post says that "this passcode-protected key material is encrypted to a Titan security chip on our datacenter floor" (emphasis added). So they have your encrypted cloud backup, and the only way to get the key material to decrypt it is to get it from an HSM in their datacenter, every part of which and the access to which they control... sounds like it's not really any better than Apple, from what I'm reading here. Granted, that article is from 2018 and I certainly have not been keeping up on android things.
[0] https://security.googleblog.com/2018/10/google-and-android-h...
HSMs are designed to protect encryption keys from everyone including the manufacturer. Signal trusts them for their encryption features. It's the best security possible for E2EE backups with passcode recovery, and Apple does it too for the subset of data that they do real E2EE backups on, like Keychain passwords. Characterizing using an HSM to implement E2EE securely as "not any better than" just giving up on E2EE for messages backups is ridiculous.
The HSMs that Signal and Apple are using are on-device though. Yes you still have to trust Signal / Apple to not exfil your key matter once decrypted by the HSM, but I submit that that is materially better than having the HSMs be hosted in a datacenter.
Signal and Apple and Google all use HSMs in datacenters as well as on device.
You can enable Advanced Data Protection to address that issue with iMessages.
Giving users an option between both paths is usually best. Most users care a lot more that they can’t restore a usable backup of their messages than they do that their messages are unreadable by the company storing them.
I used to work at a company where our products were built around encryption. Users here on HN are not the norm. You can’t trust that most users will save recovery codes, encryption seed phrases, etc in a manner that will be both available and usable when they need them, and then they tend to care a lot less about the privacy properties that provides and a lot more that they no longer have their messages with {deceased spouse, best friend, business partner, etc}.
> Apple can still read any message you exchange with practically anyone through their iCloud backups, since they are overwhelmingly likely to have backups enabled and overwhelmingly unlikely to have proactively enabled the non-default "Advanced Data Protection" feature.
> They could have implemented iMessage to not backup messages from people who enabled ADP, but they didn't. They won't even inform you when your conversation partner has uploaded your messages to Apple's servers in a form that Apple can read.
> Android's equivalent cloud backup service has been properly end-to-end encrypted by default for many years. Meaning that you don't need to convince the whole world to turn on an optional feature before your backups can be fully protected.
> Apple's stated reason for not enabling end-to-end encryption on iCloud backups by default is that it would cause data loss when users lose their devices. But Google's implementation avoids this problem. Furthermore, Apple does do end-to-end encryption by default on other critical information that would be painful to lose, such as your account passwords stored in Keychain. So that excuse doesn't seem to hold water.
This is your blog post, so I'll ask you a question. What are you trying to state in Belief #1? The message is unclear to me with how it's worded:
In addition to the things you mentioned, there's certainly a possibility of Apple attaching a virtual "shadow" device to someone's Apple ID with something like a hide_from_customer type flag, so it would be invisible to the customer.This shadow device would have it's own keys to read messages sent to your iCloud account. To my knowledge, there's nothing in the security model to prevent this.
This shadow device would have it's own keys to read messages sent to your iCloud account. To my knowledge, there's nothing in the security model to prevent this.
Matthew Green has some great posts about iMessage security. This one describes the key lookup issue:
https://blog.cryptographyengineering.com/2015/09/09/lets-tal...
Looking at the linked Apple Platform Security, it seems like the Apple Identity Service is still used as a public key directory.
Or Apple can also push an update, which you can't refuse, that upon first message to iCloud just uploads your private key. It's a bit foolish to count on encryption implemented by the adversary you're trying to hide from. Of course, this will most likely only affect individuals targeted by state-level actors.
The table has two categorizations: "In transit & on server" and "End-to-end". The former, which covers iCloud backups in the default configuration, is explicitly NOT end-to-end, meaning there are moments in time during processing where the data is not encrypted.
However, iCloud backups actually are listed as "End-to-end" if you turn on the new Advanced Data Protection feature.
262 pages!!! Pretty interesting to see how the different SoCs have evolved security wise over time.
Protects the device well... against the owner of the device using it as they wish :)
No mention of Pegasus and other software of such sort. Can latest iOS still be infected?
There is no point creating such document if elephant in the room is not addressed.
Pegasus isn't magic. It exploits security vulnerabilities just like everything else. Mitigating and fixing those vulnerabilities is a major part of this document.
Why? The obvious conclusion is that Apple is doing everything in its power to make the answer “no.”
You might as well enumerate all the viruses ever made on Windows, point to them, and then ask why Microsoft isn’t proving they’ve shut them all down yet in their documents.
That analogy misses the asymmetry in claims and power.
Microsoft does not sell Windows as a sealed, uncompromisable appliance. It assumes a hostile environment, acknowledges malware exists, and provides users and third parties with inspection, detection, and remediation tools. Compromise is part of the model.
Apple’s model is the opposite. iOS is explicitly marketed as secure because it forbids inspection, sideloading, and user control. The promise is not “we reduce risk”, it’s “this class of risk is structurally eliminated”. That makes omissions meaningful.
So when a document titled Apple Platform Security avoids acknowledging Pegasus-class attacks at all, it isn’t comparable to Microsoft not listing every Windows virus. These are not hypothetical threats. They are documented, deployed, and explicitly designed to bypass the very mechanisms Apple presents as definitive.
If Apple believes this class of attack is no longer viable, that’s worth stating. If it remains viable, that also matters, because users have no independent way to assess compromise. A vague notification that Apple “suspects” something, with no tooling or verification path, is not equivalent to a transparent security model.
The issue is not that Apple failed to enumerate exploits. It’s that the platform’s credibility rests on an absolute security narrative, while quietly excluding the one threat model that contradicts it. In other words Apple's model is good old security by obscurity.
don't worry, they set the allow_pegasus boolean to false
Apple did create a boolean for that. They call it lockdown mode.
> Lockdown Mode is an optional, extreme protection that’s designed for the very few individuals who, because of who they are or what they do, might be personally targeted by some of the most sophisticated digital threats. Most people are never targeted by attacks of this nature. When Lockdown Mode is enabled, your device won’t function like it typically does. To reduce the attack surface that potentially could be exploited by highly targeted mercenary spyware, certain apps, websites, and features are strictly limited for security and some experiences might not be available at all.
If Pegasus can break the iOS security model, there’s no reason to think it politely respects Lockdown Mode. It’s basically an admission the model failed, with features turned off so users feel like they’re doing something about it.
Lockdown mode works by reducing the surface area of possible exploits. I don't think there's any failures here. Apple puts a lot of effort into resolving web-based exploits, but they can also prevent entire classes of exploits by just blocking you from opening any URL in iMessage. It's safer, but most users wouldn't accept that trade-off.
Claiming reduced attack surface without showing which exploit classes are actually eliminated is faith, not security.
And Lockdown Mode is usually enabled _after_ user suspects targeting.
Wow, this is hardcore (pun intended).