Apple Intelligence on-device processing remains Apple’s most convincing argument in artificial intelligence. Rivals may lead in public model rankings, chatbot adoption and the speed of releasing experimental features, but Apple is building AI inside devices people already use for their most personal work.
That distinction changes the definition of winning.
Apple does not need every iPhone owner to open a separate chatbot app. It can place intelligence inside Messages, Mail, Photos, Notes, Shortcuts, Siri and third-party apps. Local models can operate on private information without automatically transferring that information to an external server.
The strategy appeared slow while the AI market measured progress through chatbot launches and model size. It looks more competitive as the industry confronts cloud costs, privacy concerns, unreliable connectivity and the difficulty of turning general-purpose AI into useful daily actions.
In his video “Apple Lost the AI Race,” Marques Brownlee captures the contradiction in one concise line:
“Apple lost the AI race. Actually, Apple won the AI race.”
The apparent reversal describes Apple’s current position well. Apple may have lost the first race for public perception. It is increasingly prepared for the longer contest over where AI runs, how it reaches users and who controls the personal data required to make it useful.
Apple Intelligence On-Device Changes the AI Relationship
Most popular generative AI products began as cloud services. A user sends a request to a remote model, the provider processes it in a data center and the answer returns through an app or browser.
That structure supports enormous models but creates dependence on internet access, server capacity and continuing infrastructure expense. It also means personal prompts leave the device, even when the provider uses strong security and privacy policies.
Apple Intelligence on-device begins from the opposite direction. The iPhone, iPad or Mac first attempts to handle the request locally using Apple Foundation Models optimized for Apple silicon.
That local model does not need to know everything available on the internet. It needs to understand the user’s request, work with relevant personal context and complete common tasks efficiently.
The difference is similar to owning a capable computer rather than renting every calculation from a distant server. Cloud systems remain useful for difficult workloads, but they no longer need to participate in every interaction.
For Apple, local processing is also a product advantage. The company controls the chip, operating system, model architecture, memory system and apps. Each layer can be designed around the others rather than assembled from unrelated services.
Image Credit: Apple Inc.
Privacy Becomes a Technical Feature
Technology companies frequently describe privacy as a policy. Apple is trying to make it part of the AI architecture.
When a request runs entirely on an iPhone, the associated data does not need to leave the device. A model can work with messages, appointments, photos or app content without uploading that personal context to a conventional cloud account.
That structure is especially relevant to Siri AI. A truly personal assistant needs access to information that users may never place willingly inside an ordinary chatbot: private conversations, travel plans, health-related details, work documents and relationships between people.
An assistant becomes more useful as it gains context. It also becomes more invasive when the same context is stored, analyzed or used for model training outside the device.
Apple Intelligence on-device allows Apple to expand personal context while maintaining a stronger boundary around it. The assistant can understand information locally, perform an action and discard temporary processing data without building a permanent advertising or training profile.
This is difficult for competitors whose businesses depend on cloud services, large-scale data collection or engagement inside separate AI products. Apple’s hardware business gives it different incentives.
Private Cloud Compute Extends the Model
Not every AI task can fit within the power and memory limits of an iPhone. Apple uses Private Cloud Compute when a request requires a larger model.
Private Cloud Compute is designed to extend the privacy architecture of Apple devices into the cloud. User information is processed only for the request, is not stored and is not made available to Apple for unrelated purposes.
Independent researchers can inspect software images used by the system, creating an unusual level of outside visibility into the cloud environment. Apple has also expanded the architecture while maintaining control over the software and privacy protections used for processing.
This creates a layered system rather than a simple choice between local and cloud AI.
Apple Intelligence first determines whether the device can handle the request. More complex work can move to Private Cloud Compute. External models may be offered separately when the user chooses to involve them.
The user does not need to understand the technical path behind every answer. The system chooses the smallest and most private environment capable of completing the task.
Apple Silicon Was an AI Investment
Apple’s control of its processors has become one of its largest advantages in the AI era.
The Neural Engine first appeared years before generative AI became a consumer obsession. It supported Face ID, photography, speech recognition and other machine-learning tasks. Apple continued expanding that hardware across the iPhone, iPad and Mac.
Those investments now provide a large installed base of devices designed to run machine-learning workloads locally. Recent Apple chips combine CPU, GPU, Neural Engine and unified memory in one architecture that can move information efficiently between different types of processing.
The same design helps foundation models run without requiring the power consumption or separate memory systems associated with conventional data-center hardware.
Apple also controls the operating systems that schedule those workloads. A model can be optimized for the exact processors, memory limits and energy behavior of supported devices.
Competitors can build excellent models, but they often need to support hardware from several manufacturers with inconsistent capabilities. Apple can decide that an AI feature requires a specific generation of silicon and design the complete experience around it.
Developers Gain a Local Model
The Foundation Models framework extends Apple Intelligence on-device beyond Apple’s own apps. Developers can use the system model through native Swift tools rather than shipping a separate large model or paying for every cloud request.
An app can generate structured content, interpret natural language or use tool calling while keeping supported processing on the device. Apple handles model delivery and system integration.
This can reduce operating costs for developers. A cloud-based AI feature creates an ongoing expense each time a user submits a request. Local inference uses hardware the customer already owns.
It can also improve reliability. Features may remain available without an internet connection and respond without waiting for a remote service.
The framework encourages a quieter form of AI adoption. Apps do not need to add a prominent chatbot tab or redesign themselves around a conversational interface. Intelligence can appear inside existing workflows as a useful capability.
That is where Apple’s strategy may reach scale. Millions of apps can incorporate small AI functions that users experience as normal software behavior rather than as visits to an AI destination.
Image Credit: Apple Inc.
The Installed Base Is Apple’s Distribution Advantage
The AI race is often discussed as a contest between models. Consumer adoption also depends on distribution.
Apple can deliver new intelligence features through operating-system updates across the iPhone, iPad, Mac and Vision Pro. Users do not need to create another account, choose a provider or transfer their personal information into a new service.
Siri already occupies the system interface. Apple Intelligence can work across first-party apps. Developers can connect their own features through Apple frameworks.
The scale is difficult to reproduce. A model provider may have stronger benchmark scores but still need users to download an app and consciously open it. Apple can introduce AI during actions people already perform: replying to a message, locating a photo, organizing a note or completing a shortcut.
This does not guarantee success. Apple must make Siri AI reliable, keep feature availability consistent and prove that local models can deliver enough capability to satisfy users accustomed to powerful cloud assistants.
The distribution advantage only becomes meaningful when the experience works.
Winning Without Building the Loudest Chatbot
Apple has made mistakes during its AI transition. Siri improvements arrived later than promised, regional regulations delayed features and competitors established powerful consumer brands before Apple completed its new system.
Those setbacks explain why the claim that Apple lost the AI race became persuasive.
They do not settle the outcome.
The first stage rewarded companies that released chatbots quickly and demonstrated increasingly large models. The next stage may reward systems that can integrate AI into ordinary life without demanding constant cloud access, high subscription costs or broad collection of personal information.
Apple Intelligence on-device is built for that stage. It uses hardware already in the user’s hand, protects more personal context locally and gives developers access to intelligence without requiring every interaction to become a server transaction.
Apple does not need to win every benchmark or release the most talkative assistant. It needs AI to become a reliable part of the devices people already trust.
Brownlee’s title works because both statements can be true. Apple lost the visible sprint that defined the early generative AI market. By concentrating on local processing, silicon, privacy and operating-system integration, it may already be positioning itself to win the race that follows.
EU tech regulation, Digital Markets Act, Apple Europe sales, Apple regulation, European Union, Siri AI, iPhone features, device interoperability, App Store changes, alternative marketplaces, iPhone Mirroring, AirPods translation, Apple privacy, iPhone security, app sideloading, platform competition, DMA compliance, Apple Intelligence, European technology, government control
EU tech regulation is no longer limited to App Store payment rules or competition disputes hidden from ordinary consumers. It is beginning to determine which features reach European devices, how quickly they arrive and how deeply third-party companies can interact with the iPhone, iPad and Mac.
The Digital Markets Act was designed to reduce the control of large technology platforms over software distribution, payments, browsers, digital assistants and connected hardware. For Apple, compliance has already produced alternative app marketplaces, web distribution, third-party payment options, additional default-app controls and expanded access to iPhone technologies.
The next phase reaches much deeper into the operating system. European regulators want competing accessories, cloud services and AI assistants to receive access to capabilities traditionally reserved for Apple’s own products.
Apple argues that some of those demands could weaken privacy and security. The European Commission says the company is responsible for creating compliant systems that preserve both competition and user protection.
Between those positions sits a large and commercially significant group of customers. Europe generated $28.06 billion in Apple sales during the quarter ending March 28, 2026, up from $24.45 billion one year earlier. Six-month European sales reached $66.20 billion, compared with $58.32 billion during the same period of fiscal 2025.
Europe therefore represented more than one-quarter of Apple’s latest quarterly revenue. It is too valuable to leave behind, but increasingly difficult to serve with the same software experience available elsewhere.
EU Tech Regulation Is Splitting Feature Availability
Regional differences once centered mainly on language support, licensing agreements or local services. EU tech regulation is creating a more structural divide.
Apple has blamed Digital Markets Act requirements for delaying Siri AI across the European Union. The company says regulators expect competing AI assistants to receive extensive access to personal information, apps and device actions before Apple can reserve those capabilities for its own assistant.
The disagreement concerns more than answering questions. Siri AI is designed to read personal context, understand onscreen information and complete actions across apps. That may include finding messages, accessing files, making purchases or sending information on the user’s behalf.
Opening those capabilities to outside AI systems requires new permission structures, technical interfaces and safeguards. Apple says it cannot release its assistant first and add competing access later without risking regulatory action. The Commission says Apple had sufficient time to design an interoperable solution.
Other delayed European features have included iPhone Mirroring on Mac, Live Translation through AirPods and certain location-based Maps capabilities. Each feature touches a different part of the ecosystem, but all depend on privileged communication between Apple devices, software and services.
The consequence is becoming visible at purchase. Two customers can buy the same iPhone model at similar prices and receive different capabilities because one lives inside the European Union.
Image Credit: Apple Inc.
Feature Adoption Becomes Slower and Less Predictable
Delayed availability can weaken the adoption of new hardware features even when the devices themselves continue selling well.
Siri AI is intended to become a central part of newer iPhones, iPads and Macs. Buyers who cannot access it lose part of the justification for newer chips, additional memory and on-device AI processing. Developers also face a smaller addressable audience for apps built around the new assistant.
AirPods Live Translation presents a similar problem. A feature designed to demonstrate the combination of iPhone processing, microphones and wearable audio becomes unavailable in a region built around dozens of languages and frequent cross-border travel.
iPhone Mirroring affects the relationship between iPhone and Mac. Users outside the EU can interact with the phone from macOS, while European users may receive notifications without gaining the complete experience.
These differences do not necessarily stop hardware sales. Apple’s newest results show strong European demand despite years of regulatory conflict. They can, however, reduce the speed at which customers use the ecosystem’s newest capabilities.
The effect may grow as software becomes a larger part of hardware differentiation. A faster processor has limited emotional value when the feature designed to demonstrate it is unavailable.
Apple Sales Show Why Europe Cannot Be Ignored
Apple does not report European Union revenue separately. Its Europe segment also includes countries outside the bloc, along with India, the Middle East and Africa. The official segment remains the best regularly published measure of the region’s commercial weight.
European sales rose approximately 15% year over year in Apple’s latest reported quarter. During the first six months of fiscal 2026, the segment expanded by roughly 14%.
Those figures weaken the argument that regulation is already driving customers away from Apple hardware. Demand remains strong, supported by the iPhone 17 family, Mac sales, services and the company’s installed base.
The financial risk is more gradual. Regional software fragmentation can raise engineering costs, delay launches and make European devices less competitive against products able to offer complete feature sets.
Apple must maintain separate distribution rules, payment systems, browser options and eligibility checks. Developers must decide whether alternative stores, external payments or EU-specific business terms justify additional work. Support teams must explain why features differ between countries using the same operating-system version.
A market responsible for more than $28 billion in one quarter can absorb considerable compliance expense. It can also influence Apple’s global product decisions when regulators demand interfaces that cannot be restricted to Europe easily.
A Slow Transfer of Ecosystem Control
The escalating regulatory environment also creates a more fundamental concern: control over complex technology ecosystems can gradually move away from the companies that designed them and toward government committees, regulatory bodies and administrative processes.
That transfer rarely happens through one dramatic order. It develops through repeated obligations covering app distribution, payment systems, browser engines, digital assistants, hardware access, data portability and interoperability. Each requirement may appear narrow when considered alone. Together, they can redefine who determines how the ecosystem operates.
Apple built its platform around tight integration between hardware, software, services and security. Regulators increasingly treat parts of that integration as market power that must be opened to competitors. The danger is that an ecosystem designed as one coordinated environment becomes subject to decisions made by committees that do not carry direct responsibility for product quality, reliability or long-term maintenance.
This can become a slow trap. Once a platform is required to expose one layer, regulators may use that access as the basis for demanding another. A payment exception can be followed by alternative app stores. Alternative stores can lead to broader installation rights. Device interoperability can expand into access for AI assistants, personal data and system actions.
The boundaries of the platform are then set less by engineering judgment and more by regulatory negotiation.
Government agencies have legitimate authority to address competition and consumer harm. The concern appears when regulation shifts from preventing abusive conduct to directing product architecture. At that point, public authorities begin influencing feature design, release timing, software permissions and the relationship between first-party and third-party services.
That creates consequences beyond Apple. Any company considering a tightly integrated product may decide that deep integration invites intervention. Businesses could avoid building ambitious ecosystems if successful coordination later becomes grounds for forced access.
Innovation then becomes shaped by what regulators are likely to permit rather than by what engineers believe will produce the safest or most coherent experience.
Image Credit: Ink Drop / Alamy Stock Photo
Competition Gains Come With Real Consumer Choices
EU policy has produced benefits that Apple rarely emphasizes. Users can install apps through alternative marketplaces or directly from authorized developers. Developers can offer external payment methods, promote different prices and avoid relying entirely on Apple’s distribution system.
Alternative browser engines can compete with WebKit, while users have more control over default apps. Contactless payment access has expanded beyond Apple Pay, allowing competing wallet providers to use the iPhone’s NFC technology under approved conditions.
Interoperability requirements could also improve third-party accessories. Smartwatches, headphones and other devices may gain better notification access, background connectivity and proximity pairing with the iPhone.
These changes can increase competition and reduce the advantage Apple gives its own services and accessories. A third-party smartwatch should not become artificially less useful simply because it was not made in Cupertino.
The difficulty is separating legitimate platform access from unlimited access. A notification interface is one thing. An autonomous AI assistant capable of reading messages, accessing files and acting across apps introduces a different level of sensitivity.
Privacy and Security Risks Are Not Theoretical
Alternative app distribution increases user choice, but it also moves some responsibility away from the App Store. Apple continues to notarize apps distributed through approved alternative channels, checking for known malware and severe fraud. That process is narrower than a complete App Store review.
Purchases made outside Apple’s payment system also fall outside its normal refund and billing support. Users may need to resolve disputes directly with the developer or payment provider.
A malicious marketplace, deceptive payment page or compromised developer account can create opportunities for fraud. Enterprise users face additional concerns when employees install software from sources their organization has not approved.
The AI question reaches further. A third-party assistant may need permission to access messages, photos, calendars, documents and app controls. Even a reputable provider can collect more information than the user expects or introduce vulnerabilities through poorly designed integrations.
Apple can build permission prompts, app isolation and revocation controls, but technical access always creates some exposure. Security depends on how narrowly permissions are defined, how visibly they are used and whether the operating system can stop unauthorized actions.
The Commission maintains that interoperability and privacy can coexist. Apple maintains that regulators are demanding access before adequate protections are ready. Both positions contain a legitimate concern: dominant platforms should not block competition, and competition should not require users to surrender control of personal data.
When regulators require a company to alter its architecture, responsibility can become blurred.
Apple remains the company customers blame when an iPhone feature fails, a third-party marketplace distributes a harmful app or an outside assistant misuses personal data. Regulators may have required the access, but they do not operate the device, provide technical support or repair the damage to the platform’s reputation.
This creates an accountability gap. Government committees can mandate openness without assuming the commercial and security consequences of that openness. Apple must comply, maintain the system and explain the outcome to users.
The same problem applies to delayed features. Regulators may argue that Apple chose not to release a service, while Apple may argue that the required conditions made the launch unsafe or technically impractical. Customers are left with less functionality and no simple way to determine who caused the delay.
Over time, the ecosystem can become a negotiated compromise rather than a product with one coherent direction. That may satisfy formal compliance while producing a weaker experience for everyone involved.
The risk is not that governments will literally operate Apple’s software. It is that committees increasingly define the conditions under which Apple can design, connect and release its own features. Control shifts through constraints.
The Industry Will Follow Europe’s Experiment
The consequences extend beyond one company. Google, Meta, Microsoft, Amazon and other platform operators face related European obligations. Hardware makers must consider whether ecosystem advantages can survive when core services are required to support rivals.
Smaller developers may gain new distribution and payment choices, but they may also inherit more compliance, customer-service and security responsibilities. Accessory companies could compete more effectively while becoming accountable for handling sensitive device data.
Other governments are watching. Rules introduced in Europe can influence legislation elsewhere, particularly when companies decide that maintaining one global technical framework is easier than operating several regional versions.
Apple is unlikely to abandon Europe or stop adapting its products. Its sales provide too much evidence of the region’s importance. The company will continue negotiating, appealing decisions and building compliance systems while withholding features it considers unsafe under current demands.
European customers may receive more app stores, payment choices and accessory compatibility. They may also wait longer for headline features included with the same hardware elsewhere.
The deeper consequence may be a technology industry that becomes more cautious about building tightly integrated products. Companies may design around anticipated regulatory demands from the beginning, limiting exclusive features, reducing private-system integration or avoiding innovations that depend on controlled access.
EU tech regulation began as an effort to constrain dominant platforms. Its long-term effect may be to transfer part of the power to shape those platforms from companies and users to regulatory institutions.
That shift deserves scrutiny equal to the market power it is intended to correct. Competition rules can restrain abuse. They can also become a slow mechanism for placing product design, security choices and ecosystem evolution under government supervision.
Shady Control and Bureaucracy
The danger grows when regulatory discretion becomes buried beneath permits, approvals, negotiations and technical committees that operate beyond meaningful public scrutiny.
The more power officials hold over which features, contracts or technologies may proceed—and when—the greater the opportunity for lobbying, favoritism and selective enforcement to influence outcomes. What begins as oversight can gradually become a system in which delay itself is leverage, while responsibility disappears inside layers of administration.
For the technology industry, that risks replacing transparent product decisions with political access, negotiated exceptions and regulatory influence exercised far from the consumers expected to live with the consequences.
The City Attorney’s Office sent the tech giants cease-and-desist letters this week telling them to stop profiting from 13 “face-swap” apps that are overwhelmingly used to target women and girls.
PrismML AI compression could give Apple a way to place far more capable artificial intelligence directly inside the iPhone. The Caltech spinout says it reduced a 27-billion-parameter model from approximately 54GB to less than 4GB, small enough to run locally on an iPhone 17 Pro.
That represents a reduction of about 93%, but the more useful number may be memory efficiency. PrismML says its technology can use up to 15 times less memory than conventional versions of the same model. Apple has reportedly held early discussions with the startup about possible uses for the technology.
Neither company has announced a partnership, acquisition or product plan. PrismML’s claims also need to be judged through independent testing, sustained performance and the compromises introduced by heavy compression. Even so, the demonstration points directly at one of Apple’s largest AI challenges: fitting stronger models inside devices without destroying battery life, consuming too much memory or sending every complex request to the cloud.
PrismML AI Compression Changes the Size Problem
Large AI models are difficult to run on phones because their parameters require substantial storage and working memory. A standard model may perform well in a data center equipped with powerful accelerators, large memory pools and constant electricity, then become impractical inside a device designed to last all day on one battery.
PrismML approaches that problem by representing model weights with far fewer bits. The startup has developed what it calls Bonsai models, including 1-bit versions designed to preserve much of the original model’s capability while sharply reducing size and energy use.
The reported iPhone demonstration used a compressed version of Alibaba’s Qwen model with 27 billion parameters. That is far larger than the compact language models normally associated with smartphones. Running it locally suggests that model size may no longer be controlled only by the number of parameters. How efficiently those parameters are represented becomes just as relevant.
Compression is not new. Developers already use quantization, pruning and other methods to make models smaller. PrismML’s claim is that its approach pushes that process much further while retaining useful reasoning and language performance.
That claim is exactly why Apple would be interested.
Why Siri AI Needs Smaller Models
Apple’s AI strategy depends on dividing work between the device and the cloud. Simpler Apple Intelligence requests can run locally, while more demanding tasks may move to Private Cloud Compute. That hybrid structure allows Apple to offer stronger capabilities without forcing the iPhone to carry every model it may ever need.
The cloud layer, however, introduces cost, latency and availability requirements. A request must leave the device, reach a server, be processed and return. Apple has designed Private Cloud Compute around strong privacy protections, but local processing remains faster and easier to explain.
A larger on-device model could improve Siri AI’s language understanding, reasoning and ability to complete multi-step tasks without a network connection. It could also reduce the number of requests sent to Apple’s servers.
That does not mean cloud processing would disappear. Server models can remain larger and support especially demanding requests. PrismML AI compression could instead move the dividing line. Tasks that currently need Private Cloud Compute might become possible directly on a future iPhone.
The benefits could include quicker responses, stronger offline operation and less server demand as Apple Intelligence reaches more users.
Image Credit: Apple Inc.
Privacy Fits Apple’s Preferred Direction
On-device processing has always been central to Apple’s AI positioning. Personal information can remain on the iPhone rather than moving through a conventional cloud service. The device can interpret a request, use local context and produce an answer without exposing the underlying data beyond the hardware.
A model capable of handling more work locally would reinforce that approach. Siri AI could potentially understand messages, notes, schedules or onscreen content while reducing the need to transfer information elsewhere.
Apple already uses Private Cloud Compute when local processing is insufficient, and the company has built significant safeguards around that system. Yet every task completed on the iPhone removes one additional processing step.
Privacy is not the only advantage. Local models can continue operating when connectivity is poor or unavailable. That could help Siri AI during flights, rural travel, crowded events or other situations where cloud access becomes unreliable.
A capable offline assistant would also feel more like part of the operating system and less like a remote service accessed through an interface.
The iPhone Still Has Physical Limits
Shrinking a model to less than 4GB does not automatically make it ideal for everyday iPhone use. The model must also run quickly enough, avoid excessive heat and leave enough memory for iOS and active apps.
A technical demonstration may complete tasks successfully while consuming more energy or time than users would accept from Siri. Apple would need to optimize the model for its Neural Engine, memory architecture and software frameworks rather than simply loading PrismML’s existing build onto every supported device.
Storage is another consideration. A model occupying several gigabytes would compete with apps, photos, offline media and system files. Apple could reserve more storage for intelligence features, deliver models selectively or use several specialized models instead of one large general system.
Battery use may create the hardest constraint. An AI model can fit in memory and still require too much sustained computation for frequent mobile use. The ideal system needs to be small, fast and efficient enough to disappear into normal iPhone behavior.
Apple’s chip design gives it an advantage here. The company controls the processor, Neural Engine, operating system and AI software stack. PrismML could provide the compression technique while Apple handles the deeper hardware integration.
Image Credit: Apple Inc.
Better AI Without Replacing the Cloud
The strongest use of PrismML’s technology may not be putting one enormous model on every iPhone. Apple could use compression across a family of models designed for different devices and tasks.
A smaller model could handle routine language requests. A more capable local model could be downloaded only to recent Pro devices with sufficient memory. Specialized models could support writing, image generation, app actions or Visual Intelligence. Private Cloud Compute could remain available when the request exceeds local capability.
That would give Apple more flexibility than an all-local or all-cloud design. Each device could carry the intelligence appropriate for its chip, memory and battery capacity.
The approach could also help Macs, iPads and future home products. A compressed model may run more comfortably on a MacBook Air, HomePod or smart home hub without requiring constant server access. Apple could scale the same intelligence foundation across products while adjusting model size to the hardware.
This is especially relevant as Siri AI moves beyond answering questions. App actions, personal context and ambient smart home control require intelligence that remains available throughout the day. Lower memory and energy requirements could make those experiences more practical.
Compression Always Includes Trade-Offs
A 93% reduction is a striking headline, but compression can affect output quality. A smaller representation may lose precision, weaken reasoning or struggle with uncommon information. The real test is not whether the model runs. It is whether it performs well enough across the tasks Apple expects Siri AI to handle.
PrismML says its technology preserves strong performance compared with full-precision models. Apple would still need extensive evaluation across languages, safety categories, tool use and personal-assistant scenarios.
Reliability is especially sensitive for an operating-system assistant. A model used for casual brainstorming can tolerate occasional weakness. A model handling messages, appointments, directions or app actions needs a higher level of consistency.
Apple may therefore use aggressive compression selectively. Some tasks can accept a smaller model, while others may remain on a higher-quality server model. The user would not need to know which model is active as long as the response remains fast and dependable.
Image Credit: Apple Inc.
Apple Wants the Technology, Not Necessarily the Company
Reports indicate Apple has spoken with PrismML, but early conversations can lead in several directions. Apple could license the startup’s technology, work on a technical integration, hire members of the team or explore an acquisition.
The company has a long history of purchasing smaller firms for specialized technology and engineering talent. Model compression would fit that pattern because it supports Apple’s existing AI strategy rather than creating a separate consumer product.
PrismML could also remain independent and provide its tools to multiple hardware companies. Its technology has potential uses across phones, computers, robotics, vehicles and industrial equipment where cloud access is expensive or unreliable.
Apple’s interest alone gives the startup attention. The larger significance is that efficient models are becoming as valuable as larger ones. The AI industry spent years treating scale as the primary route to improvement. Mobile computing forces a different question: how much intelligence can fit inside the power and memory limits of one personal device?
A Smaller Model Could Make Siri Feel Larger
Apple does not need to place the biggest available model on an iPhone. It needs a model capable enough to improve the interactions users repeat every day.
PrismML AI compression could help Siri understand more context, complete more requests locally and respond without waiting for a server. It could strengthen Apple’s privacy argument while reducing the infrastructure load created by millions of AI-enabled devices.
The technology remains early, and a startup demonstration is not the same as a shipping Apple feature. The 93% figure will mean little if quality, battery life or speed suffer under daily use.
But the direction fits Apple unusually well. The company has spent years designing smaller, more efficient hardware systems that perform beyond their apparent limits. PrismML is proposing a similar strategy for AI models.
If the startup’s claims hold up, the next major Siri improvement may not come from sending a larger model into the cloud. It may come from fitting one into the iPhone without making the iPhone feel the weight.
We live in a modern, jet-set, hyper-fast world! When we want to buy something online, boom, zoom, we use our fingerprint or face to approve the transaction, so we can grab the next Segway outta here! We don’t have time to enter a credit card! And can we trust a webpage form? Pfeh!
All right, calm down, 1950s inner voice, it’s not that bad. Most of our transactions involving a payment card or other sensitive data can be safely handled over a secure web connection. Apple Pay in Safari is the highest standard, of course, because the payment process involves encrypted elements, and your card number isn’t disclosed to the merchant. The Wallet app in iOS and the Wallet features in iPadOS and macOS further let us automate the entry of numbers and identifiers on pages we trust.
That’s all for automated commerce. What about other scenarios where you need to provide information to someone, often a friend or a local business, in order to transfer money or conduct a transaction? How can you be sure no one else is snooping in?
Make a call
Voice is still one of the most secure means of providing data. The landline wired and cellular wireless networks may be fertile ground for government agencies, but if you need to read a credit-card number or provide a PIN, a call is often the safest way to do so. By voice, you can always have the other person verify details that you believe only they know, or you can verify details to them for the same.
Because we live in a cyberpunk dystopian future, I do have to add a proviso. AI-generated voices have been convincing for a couple of years now—long enough that you shouldn’t trust an incoming call from someone you think you know, and certainly not from a bank, credit-card company, or other financial institution. I mean, I wrote “How To Avoid AI Voice Impersonation and Similar Scams” in January 2024! I assume the state-of-the-art scam is even better.
(The biggest takeaway is analog, too. Set a family password that you demand from a family member who calls asking for money or to do some weird phone or computer set of instructions.1)
Unless recorded, voice calls also lack persistence, making them impossible to recover later. A real-time Bob needs to be either between Alice and Carol or doing an AI impression of Alice or Carol.
Use secure messaging
I could make the argument that using secure messaging carries even less risk than a voice call if you’ve had an ongoing messaging conversation with someone, so you know it’s really them. Are unsecured text messages being intercepted willy-nilly?2 No, but there is also something about sending plain text all over the cellular networks that gives me the willies, nillies aside.
Look for a lock and the label Encrypted for RCS; iMessage is always encrypted, or it won’t work. (Lower image: Apple)
Secure messaging systems include:
iMessage: While iMessage has seen exploits, they’re government-grade ones. Its secure end-to-end infrastructure is creaky, but unbroken. There’s an extra detail I’ll mention below.
RCS with encryption: Between Apple Messages on iOS 26.5 or later and an Android device with RCS encryption support, you have a similar level of end-to-end protection. Make sure you see RCS encryption in the text field, or it’s just plain text. You can also use RCS encryption between two capable Android devices—as if!
Signal: The Signal app and ecosystem have the best end-to-end encryption, as it implements some of the most modern approaches to protecting older messages and current conversations available so far.
The extra note on iMessage is complicated. Briefly, if you use Messages in iCloud, an encryption key to retrieve those messages is included in an iCloud-based iPhone or iPad backup. With access to your Apple Account, someone could potentially retrieve your stored messages. However, if you enable iCloud’s Advanced Data Protection, the backup is encrypted and requires endpoint decryption using one of your devices, protecting the Messages in iCloud encryption key. This should be a worry only if someone manages to obtain your Apple Account, can activate a second factor, and the information you passed via Messages is so sensitive that someone would hunt to uncover it.
Avoid using SMS or MMS text messages (unencrypted plain text), unencrypted RCS (in-transit/at-rest encryption), or other messaging systems where you’re not sure how they handle data protection. WhatsApp can be used if you either have backups disabled or have enabled end-to-end encrypted backups.
Don’t use most collaborative tools
I know that I spend a lot of time in shared documents, whether Google Docs, Notes, Pages, or more esoteric apps or web apps. Most lack the highest level of protection. Shared Pages and Numbers files on iCloud Drive, as well as Google Docs, are encrypted in transit and at rest, and granting shared access is relatively easy (very easy in Google Docs). For iCloud Drive documents, this is true even with Advanced Data Protection enabled.
I’d be dubious about pasting my driver’s license or credit-card information into any of them that are shared with someone else. Ditto, don’t take a picture of your payment card or driver’s license and share it via a Shared Album in Photos, as it has the same issue.
There are two notable (heh) exceptions: if you and the other party or parties to a shared Notes entry or an iCloud Shared Photo Library all have Advanced Data Protection enabled, end-to-end encryption is used.
Consider the risk
As with all decisions around privacy, consider how at risk you and your data are, including voice as data. In most cases, ensuring a baseline level of encryption can be enough. Even when the app or server can “see” your unscrambled information, the transport between your device and the server and another person’s device is encrypted. Someone would have to break into the server to sniff data. If stored on the server, the baseline state would be encrypted at rest; however, the server operator would manage those encryption keys.
For peace of mind, if not strictly necessary, I’d encourage you to consider using end-to-end encryption when you can. It’s so easily available. For someone to crack your connection, they would need to obtain one of your devices and be able to unlock it. Otherwise, the path between you and someone else is effectively impregnable.
For further reading
I cover message security extensively in my book Take Control of FaceTime and Messages, which I recently updated to include the beta release of RCS encryption for Apple-to-Android communication.
For more on Wallet, I wrote an entire book on that seemingly simple app and set of features, Take Control of Apple Wallet. The book arose from the frustrations of finding where things in Wallet (and outside Wallet but related to it) lived.
On the security side of things, Take Control of Securing Your Apple Devices includes detailed advice on the ins and outs of physical security: how to ensure someone with access to your hardware can’t reach your data.
[Got a question for the column? You can email glenn@sixcolors.com or use/glennin our subscriber-only Discord community.]
This will also deter family members from asking you for money if you don’t give them the password. ↩
The United States is having something of a national referendum on mass surveillance as of late. Embodied most infamously by video surveillance company Flock Safety, Americans from all walks of life are waging war against the proliferation of the company’s automatic license plate readers (ALPR), which the ACLU warns now number in the tens of thousands.
Plenty of communities are organizing campaigns to out-muscle municipal leaders or their cheerleaders on the local police force, and by-and-large, they’re working. Whether in rural towns or sprawling cities, activists who can mobilize citizens to flood council meetings and hold picket signs are finding success. Perhaps most prominently, the Los Angeles Police Department recently allowed its three-year contract with Flock to expire, citing democratic pushback as a significant pressure point.
There’s also another, less-legal side to the anti-Flock sentiment: a rising tide of vandalism. Not content to wait for local officials to strike down ALPR contracts, concerned citizens are taking matters into their own hands by disabling or cutting down Flock camera towers wherever they’re found. But what started as a collection of separate individuals is now rising to the level of a viral phenomenon, as privacy-minded Redditors take to the web to celebrate those who are fighting back with an outpouring of ironic memes and hype reels.
Ground zero for this community appears to be the subreddit r/FlockSurveillance. The forum hosts some 422,000 “activists,” who flood the channel with over 11,000 contributions per week.
Those who’ve been caught and charged with vandalizing ALPRs become digital folk heroes on the subreddit. When news broke that US Air Force engineer Jeffrey Sovern had been charged in connection to the destruction of 13 Flock devices, redditors flooded the post with praise, sharing links to his legal defense fund and encouraging others to donate. (Sovern’s GoFundMe now stands at over $40,000, compared to around $15,440 in early July.)
“I donated to him because those charges are bulls***,” one redditor replied.
In addition to memes, the subreddit has become a destination for serious discussion around ALPRs and surveillance in general. One developer used the forum as a launch pad for their app FlockHopper, a navigation app that helps travelers avoid ALRPs. Others use it to share important developments relevant to the anti-Flock crowd, to get advice on Freedom of Information Act requests, and to dispel misinformation.
“As entertaining a trend as it is, will a few Redditors knocking a few poles over really make a difference?” one redditor asked the group.
“Even a single grain of rice can tip the scales,” a user replied.
Since WIRED reported on Meta’s NameTag face recognition system, company executives have made confusing and conflicting remarks about its very existence.
Rarely has the Los Angeles Police Department found a police toy it doesn’t love. The third-largest police force in the US has major Palantir contracts in place, uses sketchy apps to track minority communities, and routinely sends drones to spy on peaceful protestors.
So when the LAPD ends a contract with a surveillance company because it thinks the technology is too dangerous, you know there’s a problem.
Over the weekend, the police department announced it was abandoning its contract with Flock Safety, the controversial company blanketing the United States in AI-integrated license plate cameras (ALPRs). Though its three-year contract with Flock is just about up, the LAPD says it won’t renew the partnership, citing residents’ concerns over privacy and civil rights, ABC reported.
“This contract is not being renewed because of serious concerns around civil liberties and civil rights issues, particularly around privacy and the data that is being collected from these cameras,” LAPD’s chief information officer Dean Gialamas told ABC in a statement. “The LAPD had to make a difficult decision, in this case discontinuing using Flock services until we can get those data, privacy, security and sharing concerns ironed out through a contractual relationship.”
The news comes just days after our sister publication The Drive‘s Joel Feder reported how he was suddenly swarmed by armed cops while test driving a Range Rover with his wife in Minnesota. Officers said they had been tracking him for days using Flock’s AI-integrated cameras, which had erroneously tagged his car as stolen.
After the latest Flock contract news broke, 404 Media revealed that a July 10 audit by the LAPD Office of the Inspector General caught the department’s ALPR cameras generating 161 false stolen-vehicle alerts in just two months — each one ending with officers pulling over an innocent driver. Factoring in 337 alerts which “resulted in the recovery of stolen vehicles,” the LAPD’s cameras carry an error rate of 32.3 percent, effectively giving officers a one-in-three chance at pulling an innocent person over.
“During the review period, officers acknowledged 161 alerts as accurate license plate matches; however, subsequent investigations determined the vehicles were not stolen,” the report reads. “In addition to creating an inconvenience for vehicle owners, these inaccuracies can affect individual liberty interests, erode public trust, and potentially create substantial legal and financial liability concerns.”
The OIG report notes that officers are required to “attempt to verify all records prior to conducting an investigative stop,” though it didn’t disclose how many of the 161 wrongful stops were the result of an officer’s failure to do so. Instead, investigators blamed license plate records it says were “not updated in a timely manner,” which “may” lead cops to “act on inaccurate or outdated information, increasing the risk of unnecessary enforcement actions, including vehicle stops and wrongful detentions, or a confrontation with serious consequences.”
As 404 flagged, the LAPD approaches potentially stolen vehicles as “high-risk” stops. For the LAPD, these stops require drastic shows of force, including “calling for back up, air support, and a supervisor and ordering the suspect out of their vehicle,” the report reads.
It doesn’t take a large language model to figure out that this is an incredibly dangerous scenario for Black people in Los Angeles, who account for 19.5 percent of those killed by the LAPD, despite making up only nine percent of the city’s population.
Whatever way you slice it, it seems one of the most notorious police departments in the US has finally met its match.
Flock surveillance cameras are blanketing the country, and to no one’s surprise, innocent people are being caught in their dragnet.
At our sister publication The Drive, automotive journalist Joel Feder reports how he was suddenly swarmed by armed cops while test driving a Range Rover with his wife in Minnesota. The police used four squad cars to box him in in a coordinated maneuver; when he demanded an explanation, the officers said that they had been tracking him for days using Flock’s AI-integrated cameras, which had tagged his car as stolen — erroneously, as it turned out.
It was a startling example of how far reaching the controversial surveillance system had already become, never mind how flawed its conclusions can be.
“Whether you’ve actually stolen a car or are just rolling down the road having done nothing wrong, like me, once these systems have you in their crosshairs, there’s pretty much only one way it can go,” Feder wrote. “Welcome to the future. It’s scary out here.”
While still on the scene — every on-the-run criminal’s favorite place to lie low, a Kohl’s parking lot — the cops explained that the license plates on Feder’s Range Rover had been reported stolen by a dealership in Los Angeles. This made “absolutely no sense” to Feder, since car companies are extremely thorough about who in the media they loan their cars to. One officer presented photos taken by the cameras in the Flock app showing his SUV, plate number 34 10 DTM, confirming they had the right car — at least according to Flock.
But when Feder got Jaguar Land Rover on the phone, the automaker said that the reported plate number was 34 03 DTM, not 34 10 DTM. On New Jersey plates, the middle number is written smaller than the others, and for whatever reason, the number entered into the Flock system was recorded as 34 DTM.
This apparently drove Flock’s AI tech haywire, according to Feder. It “wasn’t registering that non-standard little number when it began picking up the Range Rover around town,” he wrote. “It just saw 34 DTM in large type and started alerting the local police.”
Because Flock’s surveillance network is so sprawling, this error was now a nationwide issue. “Anywhere a police department has a partnership with Flock,” Feder wrote, “any other JLR-owned car with the same plate structure is going to get flagged as stolen.” Later, Feder discovered that the car they were looking for wasn’t stolen after all; its license plate had been accidentally misplaced during a photo shoot.
Flock has surged to national attention in the past year, as police departments across the country partner with the firm to deploy its license plate readers and video cameras that can track vehicles and pedestrians. Often taking the form of towering pylons, it’s as if Flock wants residents to know they’re being watched, and its highly visible presence has served as a lightning rod for anxieties over privacy, surveillance, and police overreach.
Residents have fiercely protested against the deployment of Flock cameras, overrunning town meetings. Others have resorted to covering the cameras with trash bags or vandalizing them. Last month, a US Air Force engineer won public sympathy after being charged with sawing down over a dozen Flock cameras, with his legal defense’s GoFundMe page quickly raising tens of thousands of dollars.
On top of being widespread, Flock cameras are hooked into a centralized network, meaning the troves of footage they collect can easily be accessed by cops with little oversight. Flock and police departments argue that the cameras will only be used to investigate serious crimes, but there’re already mounting cases of cops being arrested for using Flock to stalk women. An investigation last year revealed that ICE was tapping into Flock’s network to conduct searches.
Flock’s overreach is already so vast that, in Feder’s case, he was warned to drive straight home and not cross into the neighboring towns — where their own Flock cameras would wrongly flag his car and sic the cops on him, too.
“You’re lucky we’re in Plymouth,” the officer who warned him said. “If you were in Minneapolis, they definitely would’ve come at you with guns drawn.”
The private events group, cofounded by Peter Thiel, says a “criminal” hacker is behind a breach that exposed members’ personal details. WIRED found no evidence a break-in was needed to access the files.
An anonymous reader shared this report from the Associated Press:
Officials in Kansas City, Missouri, are preparing to equip cameras on some public buses with facial recognition software capable of identifying passengers who appear on a list of banned riders or missing persons. Supporters and opponents alike view the effort as a major litmus test for tapping the AI-powered software on a U.S. public transportation system, positioning Kansas City as the latest epicenter of a fierce debate over whether the safety benefits of artificial intelligence are worth the privacy costs.
"The idea of running face recognition on a camera that is pointed on live spaces in public is a line that until recently has never really been crossed in the last 25 years," said Jay Stanley, senior policy analyst for the Project on Speech, Privacy and Technology at the American Civil Liberties Union. The state of Missouri declined to help fund the project as expected due to concerns with the facial recognition component. Still, the city is pushing ahead with local and federal money, said Tyler Means, chief mobility and strategy officer at the Kansas City Transportation Authority. "Privacy is always a tricky thing," Means said. "We've always had cameras on our buses. It's just new technology. I think in time it'll smooth over and people will realize, 'Well, it didn't really feel any different'...."
Images captured by cameras aboard the buses would immediately be checked against any active alerts, generated when a missing person, banned rider or someone on a law enforcement watch list designated by the transportation authority is identified... After the buses return to the depot, the transportation authority would archive the regular video footage on a local server for up to five years.
The company partnering with Kansas City to run the cameras "started using live facial recognition years ago to alert nursing homes when residents left the building," according to the article, and then "brought the technology to correctional institutions and schools." But this is its first attempt at bringing its cameras onto public transportation.
The article also includes this quote from Will Owen, communications director for the Surveillance Technology Oversight Project. "City residents should not be guinea pigs for transit systems to test Silicon Valley's latest unproven, biased surveillance tech."
Age verification is consuming the Internet. From social media bans in Australia to porn restrictions in half of US states, for many having to prove their age to access websites is becoming an everyday requirement. But one of the key technologies underpinning many of these age checks is about to seep into the offline world—with potentially life-changing consequences for people having their age predicted by AI.
Starting next year, the British government is planning to introduce facial age estimation—where AI scans your face and suggests how old you are—to help determine the age of asylum seekers arriving at the United Kingdom’s border. The move is believed to be the first time that a so-called facial age estimation (FAE) system has been used in this way. Many asylum seekers arriving in the UK will not have documents proving their age, and if children are incorrectly classed as adults, they can be stripped of some legal protections and placed in adult-only detention centers.
An investigation by WIRED and Lighthouse Reports, in collaboration with The Independent, has obtained an internal UK government report detailing its tests of FAE technologies. It shows how the systems regularly mistake children for adults and appear to contain serious bias problems, which directly impact the largest group of migrants subject to age assessments in 2025, according to data from the Home Office. The investigation raises questions about the effectiveness of the technology and whether it should be deployed in such high-stakes scenarios.
Artist Morry Kolman will be livestreaming feeds of the NBA champions’ ticker-tape parade from NYC’s traffic cameras—and this time, the city’s Department of Transportation isn’t demanding he stop.
Login
