➡ Tech companies like Apple, Google, and Microsoft are using AI to scan and analyze your data, including your screen, stored media, and even your keystrokes. These AI agents are always running, even if you don’t enable features like Apple Intelligence or Google Gemini. The danger is that once your personal data is stored on these devices, it can be vulnerable to hackers or government surveillance. To avoid this, you can use older devices or switch to safer platforms like Linux or Android Open Source Project.

➡ Braxtech.net sells tools that help protect your privacy from big tech’s tracking, like the Brax Virtual Phone for anonymous calls and Braxmail for creating various email identities. They also offer BytesVPN to shield your IP address from data collectors. Join their social media site, Braxme, to connect with over 100,000 users discussing privacy matters daily.

Lately, the push has been heavy for AI-capable phones and computers. Whether it’s about the iPhone 16 or macOS with Apple Intelligence, the Pixel 9 with Gemini or Copilot Plus PCs with Windows Copilot, it’s been an overwhelming AI marketing blitz in 2025. But guess what? So far, all we’ve seen are gimmicks. Genmojis. Integration with chat GPT that already exists. Background blur. Live transcriptions. Smart editing of photos. This is why you’re buying these new expensive machines like a Copilot Plus PC, or a top-of-the-line iPhone 16, or a Pixel 9. There’s a lot of promises about what AI is supposed to be doing for you, but notice that to date, we see nothing significant.

And the vast majority of people agree that all they see is fluff. So far, no device has a true AI companion. Apple Intelligence is basically vaporware. They talked up a storm about Windows Recall, but aside from upsetting users about the usefulness of this, it just raised privacy awareness and Copilot today is nothing but an obvious rapper for chat GPT. Google’s Gemini? Same generative AI gimmicks with an LLM. So these companies are making me look really bad because last year, I said specific features of these technologies are ultra dangerous. And here we are a year later, and they have not even been turned on yet.

The average person’s reaction then to new AI technologists is indifference. And these companies will quietly launch the features I fear, and trust me, they will be launched. But currently, we have a reprieve. It would be nice if AI remained at this baby stage where it stays a benign technology. I’m okay with sticking to gimmicks, honestly. Unfortunately, that’s not where it’s headed. As I’m going to show you, there’s a heavy push towards the full-scale compatibility of an AI companion. It’s a big priority for them. But while you wait for the other shoe to drop, they’re already quietly turning on the switches.

So the threat is already here. Stay right there. Watch this short clip from Marques Brownlee’s channel that shows the vaporware called Apple Intelligence. To delete a commercial that they made and published, they had to take it back because it was literally just advertising a new feature of this big Siri update that doesn’t exist yet. It’s like looking into your phone and using context to tell you something. That straight up didn’t work. Seriously, I liked it. It’s Zach. Nobody looks like Zach. You know, I think it would be cool if Siri actually could look into your phone and into your apps and calendar and actually tell you things about it.

But it can’t. As you can see, the main feature being pushed about Apple Intelligence is that it is based on a memory of you. It has to know everything you do. I’ll get into more detail on this. By the way, that video you just saw is a retracted ad from Apple. They retracted the ad because they don’t know when this feature would be released. As I said, this is currently still vaporware. Here’s a clip I’ve shown often from Ruan Cheng’s channel where he interviews the Microsoft AI CEO. We are on a mission to create a true AI companion.

And to me, an AI companion is one that can hear what you hear. And see what you see and live life essentially alongside you. Your AI companion will be able to remember everything that you’ve talked about, session to session, understand the content of the web pages that you browse, and be able to talk to you just like I’m talking to you now. So it’s going to have this seamless, fluid, very, very smooth conversational interaction. But here’s the interesting detail you may not have seen. How long until it truly starts memorizing learning you as a user over time? Yeah, great question.

Memory is the key thing that is coming soon. Certainly not this year, but we are working on it as a number one priority because it’s super important that it remembers what your preferences are and is able to kind of reason over them and give you advice based on the fact that it knows that you really don’t like modernist, you know, furniture. You like more traditional stuff. So again, we start to see that even Microsoft co-pilot is currently vaporware, at least compared to their expected full implementation. The stated priority of all these companies is AI memory and AI memory of view.

And these retracted ads and media interviews plainly show that that is the plan. Why the delay? Stylistically, each company may be tweaking the personality of their AI companion, or maybe they’re figuring out your preferred personality for your AI companion. I’m sure this is part of the marketing to make sure you like the characters they create. Kind of like making a robot that suits you. As stated by Microsoft, they likely don’t want a syncopantic AI that just acts as your slave. Maybe they wanted more assertive. Also, I’m sure Apple wants Apple Intelligence to actually be accompanied by Siri’s voice, so Siri will represent this character.

Microsoft dumped Cortana, so I don’t know if they’ll focus on a speaking machine. The point is that all these companies are striving to come out of the gate with some unique personality that they hope you will like. And I can see the problem. The AI companion cannot sound robotic. But let’s look at the deeper need to make these AI companions truly work the way they say. The problem with this concept is that the AI needs time to build information by beginning to track you now. So while you think AI is actually just a gimmick, the little features that will provide the AI memory will start to run.

And to have memory, the device will have to observe. Like Windows Recall. Like Apple’s Media Analysis D. Like Google’s System Safety Core. I made videos of all these and I’ll explain them in more detail. While the execution of the AI interface may not be completed, those little modules I mentioned are already running and they’re already accumulating data on your devices. I’m guessing it will need to compile at least a year’s worth of user data and behavior before people start to see that this AI thing is not a gimmick. So they have to start now.

And they will start with those who invested in the AI capable devices like iPhone 16, Pixel 9s, M-Silicon MacBooks and Copilot Plus PCs. Did we ask for an AI companion? The biggest marketing problem for these big tech companies is that they assume that we want an AI companion. So they are going to force it down our throat. Now you see the true fanboys like this Rowan Chung that seems so excited about this. But in order for this to progress, we the users have to like it. And I mean the majority of us.

Otherwise, we don’t buy these expensive new devices and these big companies start to fail. And it’s a risky move because so far no company has a killer app demonstrating the life enhancing value of an AI companion. We can already do LLMs with chat GPT and Grok. Still, the main focus of an AI companion is that it is a different kind of LLM. It’s an LLM that knows every detail of your life. All your preferences, all your ideas, all your beliefs, all your friends, all your media, all your keystrokes and all your voice commands.

In other words, this is a spark-like mind meld between you and the AI companion. This inherently comes with some dangerous privacy implications because what was previously known only to you is now stored on your device. AI agents that see what you see. To make the AI create a memory view, we have to start with observation, observing what you do on device. Microsoft has already made it clear how they’re going to do this and the first iteration of an observing AI agent is Windows Recall. The main way Windows Recall will observe you is through screenshots every few seconds.

Also, if you’re interacting with the AI companion in the future, whenever you query the companion, it has to automatically capture what you see on screen in a screenshot so it knows what you will be talking about. So taking screenshots will both be a historical record as well as a current observation of what you’re doing. This technology is what I refer to as see what you see. The AI agent will automatically use computer vision to analyze what you see on screen and understand the objects and text displayed. So it will already be an AI in action but not necessarily interacting with you yet.

This is an example of AI agents working with an LLM but running in the background. I want to make it clear, Windows Recall has to use some AI like computer vision or it cannot understand what is being seen and it cannot collect historical information. And by the way, the history about you will be a text-based description of what is seen in the screenshots. The screenshots themselves could be deleted. This is the way an LLM can understand what is being said. By the way, at the moment, we are currently not aware yet of any Apple module that will be doing screenshot scanning.

But understand that Apple uses OpenAI just like Microsoft so the underlying technology is the same. In order for Apple to understand context, it has to see what you see. Thus, expect Apple and Google to implement this as well. AI agents that scan. There’s another class of AI agents that again are already using AI but instead of watching your screen, they’re watching your stored media. And I get a lot of flack from this from stupid big tech psychophants that do not question technologies that have possible alternative purposes. The two examples of these are the media analysis demodule or daemon running on Apple devices and the equivalent one on new versions of Google pixels called Google system safety core.

Forget the obscure names that are used here. You can search the internet and the purpose of these modules are pretty clear. They are scanning media. Specifically on a Mac, the media scans are actually accumulated in a specific folder and you can even ask Grok where it is. So there’s no doubt that Apple devices are doing what is called client-side scanning. Scanning your content. And the description of Google system safety core is that it performs the same kind of analysis. Although supposedly it’s about media safety. You will notice by the way that certain cybersecurity pundits will claim that these modules have been tested to not communicate with either Apple or Google to pass their scanned data.

They will state that I’m fear-mongering and that my analysis has been debunked. False people. It is such a stupid assessment because that is not the point of these modules. They are A, already using AI to assess the scanned media and document them and B, they are storing these observations. C, they are meant to be data for the AI companion. This is regardless of you turning on Apple Intelligence, Google Gemini, and Windows Copilot or not. I’ll explain the true risk of this later. AI agents that keylogged. I already know that Windows Recall actually captures keystrokes or the official term is keylogging.

Capturing what you type is already directly understandable to an LLM so it can understand your messages in SMS, iMessage, WhatsApp, Signal, Telegram, or email. It also understands everything you post in social media. Keylogging is a capability of the OS that we cannot block so you cannot stop this if the OS wants to do this. This does not require any AI capability. All that needs to be done is to capture any input and store it. I can see how the data of what I type on Windows shows up in history so I can see a direct connection there.

When it comes to Apple Intelligence and Google Gemini understand that they can easily start capturing your inputs without telling you. It can be quite a quiet affair. How does Apple perform the Apple Intelligence capability demonstrated in that pulled ad if it doesn’t capture all the input data including your SMS, iMessage, and so on? They need the data. So expect they will capture what they can as media scans can be limited. In that Apple ad, Siri supposedly knew it was Zach. But how does it know the name of the person was Zach just from media scans? It can’t.

It had to have been either transcribed or captured as text from messages on the date range specified by the user. AI agents that watch the internet. In case you didn’t know this, some modules already exist to capture incoming and outgoing traffic on the internet. For example, this is already how VPNs work by capturing internet traffic going to a tunnel that is redirected to a VPN server. Certain antivirus products like Avast already scan all your internet traffic supposedly searching for phishing attacks. But the ability to hook onto the tunnel that encapsulates all internet traffic is very easy to do as Avast already demonstrates.

All it needs is an HTTPS root certificate preloaded on the device, which are obviously already preloaded by Apple, Google and Microsoft. This is an example of a data source for an AI agent to observe your activities. Again, this can be quietly done without fanfare. And like a key logger, it can theoretically scan every webpage you visit and everything you type on every application that’s using only standard TLS encryption. Now, I don’t have specific evidence that they have turned this on. However, this capability already exists and it’s already handled, like I said, by various antivirus products in Windows.

Can AI be turned off? This is a constant question that is asked of me. Can AI be turned off? Or sometimes it is less of a question but a retort to say, I’m fear-mongering because AI can be turned off. But this belies the understanding of the architecture of the AI companion model. As I explained just now, the see-what-you-see screenshots, the media scans, the key logging, and the network scan are performed by AI agents, meaning they run independently of any user interface relying on the LLM.

So without enabling Apple Intelligence or Windows Copilot or Google Gemini, these agents are running. If you try to kill the Media Analysis D module on macOS, for example, it will auto restart so you cannot stop it. The problem for Microsoft, Apple, and Google is that for some user to enable the future AI companion, the AI companion will need to have historical data from day one. And so it is likely that they will make it difficult to turn these AI agents off. The True Threat of AI Companions The real threat of the AI companion is that once you have offloaded your brain via a Vulcan mind meld with the AI companion, then the device is awfully vulnerable.

While previously your secrets were locked into your brain, you are now purposely exposing it to your device with the safety and privacy of your data under the control of MAG. Microsoft, Apple, and Google. I didn’t add Amazon. I’m going to tell you that historically speaking, if you enable this kind of technology, then every effort will be made to capture that private data in some way. Hackers will try to access that as they’ve already done with Media Analysis D and Windows Recall data. The US government will assert their rights under the Patriot Act to access anything they want.

Hey, remember, you voted for that. So now you moved the data away from the protection of the Fourth Amendment because it can be read by the AI. The AI is not something you control. So in theory, the AI can work in reverse. Just like you can use AI to ask a question and it will use your personal data to enhance the query, then someone can do the reverse and ask the AI to find certain personal data from your device. Apple already demonstrated they can do this.

They talked about it. I know Apple, Microsoft, and Google with a combined market cap of likely 7 trillion plus tells you that they won’t do that. It is already well known that they were paid millions to supply clandestine data to the NSA as documented by the three-letter agency editors at Wikipedia. So they’re not even hiding it. My philosophy is that if the technology can be abused, then it will be abused. You do not often have such an opportunity here for data capture that can cancel out the decades of surveillance from Google with something better than Google.

Your mind-meld data. How to avoid this? This can all be avoided. You do not have to participate in a mind-meld with the device. While you can all decline to use the AI interface, likely you will not be able to stop the AI agents working in the background. But only if you’re using the OS from Apple, Microsoft, and Google. So the current problem platforms are macOS from mSilicon, iOS 15+, Windows 11, and Google Android 9+. If you’re using older Windows, macOS for Intel Macs, old iOS, and old Android, then you are not subject to this risk.

However, honestly, this would be very old devices, so they are ready for the landfill. If you’re looking at new devices, the safe devices are any computers running Linux natively, meaning not as a virtual machine. And devices based on Android Open Source Project, which includes our Lunar OS, EOD OS, Linux OS, Calix OS, Graphene OS, and eFoundation. This also includes Ubuntu Touch, which is a Linux Mobile OS. In my last video, I mainly focused on powerful laptops that can do AI, but can safely run Linux, so they will be safe for the long term.

You can run safe AI as long as it is not a true AI companion. In summary, although we have what we think is a reprieve from AI companions, the reality is that AI agents can be turned on immediately, which will begin collecting data now. The threat is already here, regardless of the supposed progress of final AI products. So you need to take action now. Folks, as many of you know, this channel does not have sponsors. Instead, we rely completely on community support to keep us going.

Hopefully, we give you enough information that you find value. Thank you for those who support us on Patreon, locals, and YouTube memberships. For most of our support, we instead strive to provide products and services that help you in your privacy journey. Our newest product is the Brax III Privacy Phone. This is a community project involving several companies and is sold on Braxtech.net. This is an important tool in the privacy battle as it is immune from things like location tracking from big tech. We have other products and services that can be found on our privacy social media site, Braxme.

We have Brax Virtual Phone that gives you inexpensive, no KYC phone numbers for better anonymity. We have Braxmail, which offers unlimited aliases and many domains to create different email identities. We have BytesVPN, which protects your IP addresses from being harvested by data brokers. Join us on Braxme and meet the way over 100,000 users who discuss privacy issues daily. Thank you for watching and see you next time. [tr:trw].

See more of Rob Braxman Tech on their Public Channel and the MPN Rob Braxman Tech channel.