Full title is "Memory Integrity Enforcement: A complete vision for memory safety in Apple devices"
This is great, and a bit of a buried lede. Some of the economics of mercenary spyware depend on chains with interchangeable parts, and countermeasures targeting that property directly are interesting.
(1) AOSP isn't dead, but Google just landed a huge blow to custom ROM developers: https://www.androidauthority.com/google-not-killing-aosp-356...
(2) Privacy-Focused GrapheneOS Warns Google Is Locking Down Android: https://cyberinsider.com/privacy-focused-grapheneos-warns-go...
(3) GrapheneOS exposes Google's empty promises on Android security updates: https://piunikaweb.com/2025/09/08/grapheneos-google-security...
Xbox One, 2012? Never hacked.
Nintendo Switch 2, 2025? According to reverse engineers... flawlessly secure microkernel and secure monitor built over the Switch 1 generation. Meanwhile NVIDIA's boot code is formally verified this time, written in the same language (ADA SPARK) used for nuclear reactors and airplanes, on a custom RISC-V chip.
iPhone? iOS 17 and 18 have never been jailbroken; now we introduce MIE.
There are still plenty of other flaws besides memory unsafety to exploit. I doubt that we'll see like a formally proven mainstream OS for a long time.
> ... Google recently made incredibly misguided changes to Android security updates. Android security patches are (now) almost entirely quarterly instead of monthly to make it easier for OEMs. They're giving OEMs 3-4 months of early access.. Google's existing system for distributing security patches to OEMs was already incredibly problematic. Extending 1 month of early access to 4 months is atrocious. This applies to all of the patches in the bulletins.
> ... The existing system should have been moving towards shorter broad disclosure of patches instead of 30 days. Moving in the opposite direction with 4 months of early access is extraordinarily irresponsible. ...Their 3-4 month embargo has an explicit exception for binary-only releases of patches. We're fully permitted to release the December 2025 patches this month in a release but not the source code.
> Nearly all OEMs were failing to ship the monthly security patch backports despite how straightforward it is. The backports alone are not even particularly complete patches. They're only the High and Critical severity Android patches and a small subset of external patches for the Linux kernel, etc. Getting the full Android patches requires the latest stable releases.
It’s my understanding that this won’t protect you in the case where the attacker has a chance to try multiple times.
The approach would be something like: go out of bounds far enough to skip the directly adjacent object, or do a use after free with a lot of grooming, so that you get a a chance of getting a matching tag. The probability of getting a matching tag is 1/16.
But this post doesn’t provide enough details for me to be super confident about what I’m saying. Time will tell! If this is successful then the remaining exploit chains will have to rely on logic bugs, which would be super painful for the bad guys
That makes the probability work against the attacker really well. But it’s not a guarantee
Apple are definitely doing the best job that any firm ever has when it comes to mitigation, by a wide margin. Yet, we still see CVEs drop that are marked as used in the wild in exploit chains, so we know someone is still at it and still succeeding.
When it comes to the Xbox One, it’s an admirable job, in no small part because many of the brightest exploit developers from the Xbox 360 scene were employed to design and build the Xbox One security model. But even still, it’s still got little rips at the seams even in public: https://xboxoneresearch.github.io/games/2024/05/15/xbox-dump...
But yeah this was support for a the longest time by IBM basically. It's nice to see it's getting more widespread.
They're not so much general purpose computers anymore as they are locked down bank terminals.
Hitting people with wrenches leaves marks that can be shown to the media and truth & reconciliation commissions. Wetwork and black-bagging dissidents leaves records: training, operational, evidence after the fact. And it hardly scales – no matter what the powers at be want you to think, I think history shows there are more Hugh Thompsons than Oskar Dirlewangers, even if it takes a few years to recognize what they've done.
If we improve security enough that our adversaries are _forced_ to break out the wrenches, that's a very meaningful improvement!
> Extensions provide no security. [...] The tagged memory extensions don't stop you from doing anything.
>With the introduction of the iPhone 17 lineup and iPhone Air, we’re excited to deliver Memory Integrity Enforcement: the industry’s first ever, comprehensive, always-on memory-safety protection covering key attack surfaces — including the kernel and over 70 userland processes — built on the Enhanced Memory Tagging Extension (EMTE) and supported by secure typed allocators and tag confidentiality protections.
Of course it is a little disappointing not to see GrapheneOS's efforts in implementing [1] and raising awareness [2] recognised by others but it is very encouraging to see Apple making a serious effort on this. Hopefully it spurs Google on to do the same in Pixel OS. It should also inspire confidence that GrapheneOS are generally among the leaders in creating a system that defends the device owner against unknown threats.
[1] https://grapheneos.org/releases#2023103000 [2] https://xcancel.com/GrapheneOS/status/1716946325277909087#m
CHERI-Morello uses 129-bit capability objects to tag operations, has a parallel capability stack, capability pointers, and requires microarchitectural support for a tag storage memory. Basically with CHERI-Morello, your memory operations also need to provide a pointer to a capability object stored in the capability store. Everything that touches memory points to your capability, which tells the processor _what_ you can do with memory and the bounds of the memory you can touch. The capability store is literally a separate bus and memory that isn't accessible by programs, so there are no secrets: even if you leak the pointer to a capability, it doesn't matter, because it's not in a place that "user code" can ever touch. This is fine in theory, but it's incredibly expensive in practice.
MIE is a much simpler notion that seems to use N-bit (maybe 4?) tags to protect heap allocations, and uses the SPTM to protect tag space from kernel compromise. If it's exactly as in the article: heap allocations get a tag. Any load/store operation to the heap needs to provide the tag that was used for their allocation in the pointer. The tag store used by the kernel allocator is protected by SPTM so you can't just dump the tags.
If you combine MIE, SPTM, and PAC, you get close-ish to CHERI, but with independent building blocks. It's less robust, but also a less granular system with less overhead.
MIE is both probabilistic (N-bits of entropy) and protected by a slightly weaker hardware protection (SPTM, which to my understanding is a bus firewall, vs. a separate bus). It also only protects heap allocations, although existing mitigations protect the stack and execution flow.
Going off of the VERY limited information in the post, my naive read is that the biggest vulnerability here will be tag collision. If you try enough times with enough heap spray, or can groom the heap repeatedly, you can probably collide a tag with however many bits of entropy are present in the system. But, because the model is synchronous, you will bus fault every time before that, unlike MTE, so you'll get caught, which is a big problem for nation-state attackers.
Yes: if you have half of a billion dollars in BTC, sure – you're a victim to the wrench, be it private or public. If you're a terrorist mastermind, you're likely going to Gitmo and will be placed in several stress positions by mean people until you say what they want to hear.
Extreme high-value targets always have been, and always will be, vulnerable to directed attacks. But these improvements are deeply significant for everyone who is not a high-value target – like me, and (possibly) you!
In my lifetime, the government has gone from "the feds can get a warrant to record me speaking, in my own voice, to anyone I dial over my phone" to "oh, he's using (e2e encrypted platform) – that's a massive amount more work if we can even break it". That means the spectrum of people who can be targeted is significantly lower than it used to be.
Spec-fiction example: consider what the NSA could do today, with whisper.cpp & no e2e encrypted calls.
PAC may stop you from changing values - or at least you'd have to run code in the process to change them.
* use synchronous exceptions (“precise-mode”), which means the faulted instruction cannot retire and cause damage
* re-tag allocations on free
FWIW, I presume this is "from experience"--rather than, from first principles, which is how it comes off--as this is NOT how their early kernel memory protections worked ;P. In 2015, with iOS 9, Apple released Kernel Patch Protection (KPP), which would verify that the kernel hadn't been modified asynchronously--and not even all that often, as I presume it was an expensive check--and panic if it detected corruption.
https://raw.githubusercontent.com/jakeajames/rootlessJB/mast...
> First let’s consider our worst enemy since iOS 9: KPP (Kernel Patch Protection). KPP keeps checking the kernel for changes every few minutes, when device isn’t busy.
> That “check every now and then” thing doesn’t sound too good for a security measure, and in fact a full bypass was released by Luca Todesco and it involves a design flaw. KPP does not prevent kernel patching; it just keeps checking for it and if one is caught, panics the kernel. However, since we can still patch, that opens up an opportunity for race conditions. If we do things fast enough and then revert, KPP won’t know anything ;)
As an outsider I am quite ignorant to what security developments these companies are considering and when the trade-offs are perhaps too compromising for them to make it to production. So I can't appreciate the scale of what Apple had to do to reach this stage, whereas with GrapheneOS I know they favour privacy/security on balance. I use that as a weak signal to gauge how committed Apple/Google/Microsoft are to realising those kinds of goals too.
They also imply a very different system architecture.
Why would you need MTE if you have CHERI?
But here’s a reason to do both: CHERI’s UAF story isn’t great. Adding MTE means you get a probabilistic story at least
I think it's two halves of the same coin and Apple chose the second half of the coin.
The two systems are largely orthogonal; I think if Apple chose to go from one to the other it will be a generational change rather than an incremental one. The advantage of MTE/MIE is you can do it incrementally by just changing the high bits the allocator supplies; CHERI requires a fundamental paradigm shift. Apple love paradigm shifts but there's no indication they're going to do one here; if they do, it will be a separate effort.
Correct me if I'm wrong, but the spyware that has been developed certainly could be applied at scale at the push of a button with basic modification. They just have chosen not to at this time. I feel like this paragraph is drawing a bigger distinction than actually exists.
Overall my _personal_ opinion is that CHERI is a huge win at a huge cost, while MTE is a huge win at a low cost. But, there are definitely vulnerability classes that each system excels at.
That’s strictly better, in theory.
(Not sure it’s practically better. You could make an argument that it’s not.)
And CHERI fixes it only optionally, if you accept having to change a lot more code
There have been multiple full-chain attacks since the introduction of PAC. It hasn’t been a meaningful attack deterrent because attackers keep finding PAC bypasses. This should give you pause as to how secure EMTE actually is.
I also think this argument is compelling because one exists in millions of consumer drives, to-be-more (MTE -> MIE) and one does not.
More interesting is how to trace and debug code on such a CPU. Because what a debugger often does is exactly patching an executable in RAM, peeks and pokes inside, etc. If such an interface exists, I wonder how is it protected; do you need extra physical wires like JTAG? If it does not, how do you even troubleshoot a program running on the target hardware?
> We have used CHERI’s ISA facilities as a foundation to build a software object-capability model supporting orders of magnitude greater compartmentalization performance, and hence granularity, than current designs. We use capabilities to build a hardware-software domain-transition mechanism and programming model suitable for safe communication between mutually distrusting software
and https://github.com/CTSRD-CHERI/cheripedia/wiki/Colocation-Tu...
> Processes are Unix' natural compartments, and a lot of existing software makes use of that model. The problem is, they are heavy-weight; communication and context switching overhead make using them for fine-grained compartmentalisation impractical. Cocalls, being fast (order of magnitude slower than a function call, order of magnitude faster than a cheapest syscall), aim to fix that problem.
This functionality revolves around two functions: cocall(2) for the caller (client) side, and coaccept(2) for the callee (service) side. Underneath they are implemented using CHERI magic in the form of CInvoke / LDPBR CPU instruction to switch protection domains without the need to enter the kernel, but from the API user point of view they mostly look like ordinary system calls and follow the same conventions, errno et al.
There's a decent chance that we get back whatever performance we pay for CHERI with interest as new systems architecture possibilities open up.
MTE helps us secure existing architectures. CHERI makes new architectures possible.
Did they ever explain what that mitigation does?
Pixels are not the only Android devices with MTE anymore and haven't been for a while. We've tried it on a Samsung tablet which we would have liked to be able to support if Samsung allowed it and did a better job with updates.
GrapheneOS is not a 1 person project and not a hobby project. I wasn't the one to implement MTE for hardened_malloc and have not done most of the work on it. The work was primarily done by Dmitry Muhomor who is the lead developer of GrapheneOS and does much more development work on the OS than I do. That has been the case for years. GrapheneOS is not my personal project.
We've done a large amount of work on it including getting bugs fixed in Linux, AOSP and many third party apps. Our users are doing very broad testing of Android apps with MTE and reporting issues to developers. There's a specific crash reporting system we integrated for it to help users provide usable information to app developers. The hard part is getting apps to deal with their memory corruption bugs and eventually Google is going to need to push for that by enabling heap MTE by default at a new target API level. Ideally stack allocation MTE would also be used but it has a much higher cost than heap MTE which Apple and Google are unlikely to want to introduce for production use.
Android apps were historically largely written in Java which means they have far fewer memory corruption bugs than desktop software and MTE is far easier to deploy than it otherwise would be. Still, there are a lot of native libraries and certain kinds of apps such as AAA games with far more native code have much bigger issues with MTE.
Sure, the whole sentence is a bit of a weird mess. Paraphrased: it made exploits more complex, so we concluded that we needed a combined SW/HW approach. What I read into that is that they're admitting PAC didn't work, so they needed to come up with a new approach and part of that approach was to accept that they couldn't do it using either SW or HW alone.
Then again... I don't know much about PAC, but to me it seems like it's a HW feature that requires SW changes to make use of it, so it's kind of HW+SW already. But that's a pointless quibble; EMTE employs a lot more coordination and covers a lot more surface, iiuc.
Perhaps the real problem is that you can use speculation to scan large amounts of memory for matching tags, some of which would be different types, so you need something to handle that?
(talking out of my butt here)
The main weakness is that MTE is only 4 bits... and it's not even 1/16 but typically 1/15 chance of bypassing it since a tag is usually reserved for metadata, free data, etc. The Linux kernel's standard implementation for in-kernel usage unnecessarily reserves more than 1 to make debugging easier. MTE clears the way for a more serious security focused memory tagging implementation with far more bits and other features. It provides a clear path to providing very strong protection against the main classes of vulnerabilities used in exploits, especially remote/proximity ones. It's a great feature but it's more what it leads to that's very impressive than the current 4 bit MTE. Getting rid of some known side channels doesn't make it into a memory safety implementation.
Qualcomm has to make their own implementation which has significantly delayed widespread availability. Exynos and MediaTek have it though.
If that's the case, then many of their public statements about this are extraordinarily dishonest. There are widespread exploits targeting Safari, Chrome, iOS and Android. These are not only rare attacks targeting people heavily sought out by governments, etc. They do not have nearly as much visibility into it as they make it seem.
For example, I might know of an unrelated exploit I'm sitting on because I don't want it fixed and so far it hasn't been.
I think the climate has become one of those "don't correct your adversary when they make mistakes" types of things versus an older culture of release clout.
Okay a bit drastic, I don’t really know if this will affect them.
There's widespread exploitation of Apple devices around the world by many governments, companies, etc. Apple and Google downplay it. The attacks are often not at all targeted but rather you visit a web page involving a specific political movement such as Catalan independence and get exploited via Safari or Chrome. That's not a highly targeted attack and is a typical example of how those exploits get deployed. The idea that they're solely used against specific individuals targeted by governments is simply not true. Apple and Google know that's the case but lead people to believe otherwise to promote their products as more safe than they are.
> I think SEAR is extremely aware of what real-world exploitation of iPhones looks like.
Doesn't seem that way based on their interactions with Citizen Lab and others.
I interpreted that as what they came up with when first looking at/starting to implement MTE, not their plan since $longTimeAgo.
Apple has certainly gotten better about security, and I suspect things like what you listed are a big part of why. They were clearly forced to learn a lot by jailbreakers.
Personally I didn’t read it as a swipe against Android. If it was I don’t personally know what attack(s) it’s referring to outside of the possibility of malware installed by the vendor.
But if it’s installed by the vendor, they can really do anything can’t they. That’s not really a security breach. Just trust.
It sounds like the kernel’s allocations may only use one tag(?). So if you get in there, jackpot right? No tags to deal with.
So they’re using special compiler flags to limit all offsets to less than 4 GB. Then they placed different parts of the kernel far apart in address space with a 4 GB unmapped zone.
So if you can put your own pointer somewhere that’s exploitable in allocated kernel memory, there is no way for it to point to any other “part” of kernel memory. Only within that one “area”.
Presumably this would mean that exploiting a problem in the graphics drivers would not make it possible to provide a pointer pointing to the Secure Enclave interface code. Or something like that.
I’m not 100% on if I’m understanding it correctly.
Nice to hear it’s already in use in some forms.
And of course it seems pretty obvious that if this is in the new iPhones it’s going to be in the M5 or M6 chips.
The 202X M-series don’t always have the same core revisions as the A-series. Sometimes they’re based on the cores from 202X-1.
Given how nice a feature it is I certainly hope it’s in the M5.
Others are aware of where MTE needs improvement and are working on it for years. Cortex shipped MTE with a side channel issue which is better than not shipping it and it will get addressed. Apple has plenty of their own side channel vulnerabilities for their CPUs. Deterministic protections provided via MTE aren't negatively impacted by the side channel and also avoid depending on only 4 bits of entropy. The obvious way to use MTE is not the only way to use it.
GrapheneOS began using MTE in production right after the Pixel 8 provided a production quality implementation, which was significantly later than it could have been made available since Pixels aren't early adopters of new Cortex cores. On those cores, asynchronous MTE is near free and asymmetric is comparable to something like -fstack-protector-strong. Synchronous is relatively expensive, so making that perform better than the early Cortex cores providing MTE seems to be where Apple made a significant improvement. Apple has higher end, larger cores than the current line of Cortex cores. Qualcomm's MTE implementation will be available soon and will be an interesting comparison. We expect Android to heavily adopt it and therefore it will be made faster out of necessity. The security advantage of synchronous over asymmetric for userspace is questionable. It's clearer within the kernel, where little CPU time is spent on an end user device. We use synchronous in the kernel and asymmetric in userspace. We haven't offered full synchronous as an option mainly because we don't have any example of it making a difference. System calls act as a synchronization point in addition to reads. io_uring isn't available beyond a few core processes, etc.
That's Apple and here is Google (who have been at memory safety since the early Chrome/Android days):
Google folks were responsible for pushing on Hardware MTE ... It originally came from the folks who also did work on ASAN, syzkaller, etc ... with the help and support of folks in Android ... ARM/etc as well.
I was the director for the teams that created/pushed on it ... So I'm very familiar with the tradeoffs.
...
Put another way - the goal was to make it possible to use have the equivalent of ASAN be flipped on and off when you want it.
Keeping it on all the time as a security mitigation was a secondary possibility, and has issues besides memory overhead.
For example, you will suddenly cause tons of user-visible crashes. But not even consistently. You will crash on phones with MTE, but not without it (which is most of them).
This is probably not the experience you want for a user.
For a developer, you would now have to force everyone to test on MTE enabled phones when there are ~1mn of them. This is not likely to make developers happy.
Are there security exploits it will mitigate? Yes, they will crash instead of be exploitable. Are there harmless bugs it will catch? Yes.
...
As an aside - It's also not obvious it's the best choice for run-time mitigation.
Google Security (ex: TAG & Project Zero) do so much to tackle CSVs but with MTE the mothership dropped the ball so hard.
What about the blogpost suggested this?
" ... always-on memory safety protection for our key attack surfaces including the kernel ..."
" ... always-on memory-safety protection covering key attack surfaces — including the kernel and over 70 userland processes — built on the Enhanced Memory Tagging Extension (EMTE) and supported by secure typed allocators and tag confidentiality protections ... "
Suggests to me that the kernel allocator uses a similar tagging policy as the userspace allocators do.
Google set it up for usage on Pixels, and then later Samsung and others did too. Pixel 8 was the first device where it was actually usable and production quality. GrapheneOS began using it in production nearly immediately after it launched on the Pixel 8.
*: or whatever else people use jailbreaks for these days
I just want to address this part. Why shouldn't Apple advertise or market its achievements here? If they're effectively mitigating and/or frustrating real world attacks and seems to eliminate a class of security bugs, why shouldn't they boast about it; it shows that security R&D is in the forefront of the products they build which is an effective strategy for selling more product to the security conscious consumer.
Not a shill, but a shareholder, and I invest in Apple because they're at the forefront of a lot of tech.
It's often external parties finding exploits being used in the wild and reporting it to Apple and Google. Citizen Lab, Amnesty International, etc.
We regularly receive info from people working at or previously working at companies developing exploits and especially from people at organization using those exploits. A lot of our perspective on it is based on having documentation on capabilities, technical documents, etc. from this over a long period of time. Sometimes we even get access to outdated exploit code. It's major releases bringing lots of code churn, replaced components and new mitigations which seem to regularly break exploits rather than security patches. A lot of the vulnerabilities keep working for years and then suddenly the component they exploited was rewritten so it doesn't work anymore. There's not as much pressure on them to develop new exploits regularly as people seem to think.
MTE is 4 bits with 16 byte granularity. There's usually at least 1 tag reserved so there are 15 random tags. It's possible to dynamically exclude tags to have extra deterministic guarantees. GrapheneOS excludes the previous random tag and adjacent random tags so there are 3 dynamically excluded tags which were themselves random.
Linux kernel MTE integration for internal usage is not very security focused and has to be replaced with a security-focused implementation integrated with pKVM at some point. Google's recently launched Advanced Protection feature currently doesn't use kernel MTE.
AOSP's security posture is frustrating (as Google seemingly solely decides what's good and what's bad and imposes that decision on each of their 3bn users & ~1m developers, despite some in the security community, like Daniel Micay, urging them to reconsider). The steps Apple has been taking (in both empowering the developers and locking down its own OS) in response to Celebgate and Pegasus hacks has been commendable.
You can fix this insofar as you control the compiler and calls to malloc(), which you don't, because third party code may have wrappers around it.
There is one stack, the normal program stack that's normal main memory.
> capability pointers
If you use pure-capability CHERI C/C++ then there is only one type of pointer to manage; they just are implemented as capabilities rather than integers. They're also just extensions of the existing integer registers; much as 64-bit systems extend 32-bit registers, CHERI capability registers extend the integer registers.
> requires microarchitectural support for a tag storage memory
Also true of MTE?
> your memory operations also need to provide a pointer to a capability object stored in the capability store
There is no "capability object stored in the capability store". The capability is just a thing that lives in main memory that you provide as your register operand to the memory instruction. Instead of `ldr x0, [x1]` to load from the address `x1` into `x0`, you do `ldr x0, [c1]` to load from the capability `c1`. But `c1` has all of the capability; there is no indirection. It sounds like you are thinking of classical capability systems that did have that kind of indirection, but an explicit design goal of CHERI is to not do that in order to be much more aligned with contemporary microarchitecture.
> The capability store is literally a separate bus and memory that isn't accessible by programs,
As above, there is no separate bus, and capabilities are not in separate memory. Everything lives in main memory and is accessed using the same bus. The only difference is there are now capability tags being stored alongside that data, with different schemes possible (wider SRAM, DRAM ECC bits, carving out a bit of main memory so the memory controller can store tags there and pretend to the rest of the system that memory itself stores tags). To anything interacting with the memory subsystem, there is one bus, and the tags flow with the data on it.
>None of this is for users
I got a bit confused when reading this. What does it mean to "know the tag" if the memory region is untagged?
This point could use more explanation. The fundamental problem here is the low entropy of the tags (only 4 bits). An attacker who randomly guesses the tags has 1/16 chance of success. That is not fixed by reseeding the PRNG. So I am not sure what they mean.
In practice, it is 15/16 chance of detection of the exploit attempt. Which is an extraordinarily high rate of detection, which will lead to a fix by Apple.
Net net, huge win. But I agree they come across as overstating the prevention aspect.
Just like any weapon, "security" is only good if it's in your control. When the noose is around your neck, you'd better hope it easily breaks.
b your iphones BEEN pwned for YEARS and it was done in minutes LOL. gtfoh
with help from ChatGPT: Apple claims “never been a successful iPhone malware attack” Reality: WireLurker, Masque, XcodeGhost, YiSpecter, jailbreak 0-days, Pegasus/Predator/Reign 0-clicks.
iPhones pwned for yrs — by kids in pajamas
There is a section in the technical reports that talks about garbage collection.
I don't think CHERI is currently being used with different privileged threads in the same address space.
To the architecture, there is one access mechanism with the tag bit set and one separate mechanism with the tag bit unset, no?
I thought this was the whole difference: in MTE, there is a secret tag hidden in a “normal” pointer by the allocator, and in CHERI, there is a separate architectural route for tag=0 (normal memory) and tag=1 (capabilities memory), whether that separate route eventually goes to some partition of main memory, a separate store entirely, ECC bit stuffing, or whatever?
Unfortunately, like in many other cases, Intel botched their MPX design, only evolutions of MTE and CHERI are around.
I do agree it is a pain not seeing this becoming widely adopted.
As for disabling JIT, it would have the same effect as early Androids, lagging behind Symbian devices, with applications that were wrappers around NDK code.
In November 2021, Apple Inc. filed a complaint against NSO Group and its parent company Q Cyber Technologies in the United States District Court for the Northern District of California in relation to FORCEDENTRY, requesting injunctive relief, compensatory damages, punitive damages, and disgorgement of profits but in 2024 asked the court to dismiss the lawsuit.
Well, Apple already routinely forces developers to recompile their applications so if Apple wants to introduce something needing a compiler / toolchain update they can do that easily. And they also control the entire SoC from start to finish and unlike pretty much everyone else also hold an ARM Architecture License so they can go and change whatever they want in the hardware side as well.
Not to mention the dynamic linker.
In MTE, you have the N-bit (typically 4) per-granule (typically 16 byte) "colour"/tag that is logically part of the memory but the exact storage details are abstracted by the implementation. In CHERI, you have the 1-bit capability tag that is logically part of the memory but the exact storage details are abstracted by the implementation. If you understand how MTE is able to store the colours to identify the different allocations in memory (the memory used for the allocations, not the pointers to the allocations) then you understand how CHERI stores the tags for its capabilities, because they are the same basic idea. The difference comes in how they're used: in MTE, they identify the allocation, which means you "paint" the whole allocation with the given "colour" at allocation time (malloc, new, alloca / stack variables, load time for globals), but in CHERI, they identify valid capabilities, and so only get set when you write a valid capability to that memory location (atomically and automatically). This leads to very different access patterns and densities (e.g. MTE must tag all data regardless of its type, whereas CHERI only tags pointers, meaning large chunks of plain data have large chunks of zero tag bits, so how you optimise your microarchitecture changes).
Perhaps you're getting confused with details about the "tag table + cache" implementation for how tags can be stored in commodity DRAM? For CHERI you really want 129-bit word (or some multiple thereof) memory, but commodity DRAM doesn't give you that. So as part of the memory controller (or just in front of it) you can put a "tag controller" which hides a small (< 1%) fraction of the memory and uses it to store the tags for the rest of the memory, with various caching tricks to make it go fast. But that is just the tag, and that is an implementation detail for how to pretend that your memory can tag data. You could equally have an implementation that uses wider DRAM (e.g. in the case of DRAM with ECC bits to spare). Both schemes have been implemented. But importantly memory is just 128+1-bit; the same 128 bits always store the data, whether it's some combination of integers and floats, or the raw bytes of a capability. In the former case, the 129th tag bit will be kept as 0, and in the latter case it will be kept as whatever the capability's tag is (hopefully 1).
My impression is that Apple's threat intelligence effort is similar in quality to Google's. Of course external parties also help but Apple also independently finds chains sometimes.
Daniel's position on MTE for a while has been that Google is dragging their feet in turning it on, but he fails to understand that there is more to it than just flipping a switch that he does in his OS. To actually productionize it requires a huge amount of effort that Apple put in here and Daniel, as talented as he is, really can't do. We know this because Google was not able to do it even though they wanted to. (For the avoidance of doubt: Google does want to turn on MTE, they're not just dawdling "just because". The current MTE implementation is not good enough for them.)
Maybe you've been confused by a description of how it works inside a processor. In early CHERI designs, capabilities were in different architectural processor registers from integers.
In recent CHERI designs, the same register numbers are used for capabilities and other registers. A micro-architecture could be designed to have either all registers be capability registers with the tag bit, or use register renaming to separate integer and capability registers.
I suppose a CHERI MCU for embedded systems with small memory could theoretically have tag pages in separate SRAM instead of caching main memory, but I have not seen that.
With CHERI, there is nothing to guess. You either have a capability or you don't.
It isn't great. Most users won't assume malice when an app crashes. And if they reopen it a few times your chance of succeeding goes up quickly. But this is also assuming that you need a single pointer tag to exploit something. If you need more you need to get even luckier.
So it definitely isn't perfect protection. But it isn't trivial to bypass.