0:00 Firefox is now the last major browser that fully supports the ad-blocker uBlock Origin.0:05 Microsoft Edge just became the latest browser to effectively disable it, following a change that Google pioneered.0:12 Last week, we talked about an AI acing a PhD exam in just eleven hours, a massive leap in capability.0:19 This week, the conversation is shifting from what these powerful systems can do, to who gets to control what they see, and what you see.0:27 It’s a battle over control, and your web browser is right on the front lines.0:32 The story starts with something called Manifest V3.0:36 It's a technical standard for browser extensions, and the important thing to know is that it severely limits the power of ad-blockers.0:44 Chrome, and all the browsers built on its code like Opera, Brave, and now Microsoft Edge, have all moved to it.0:51 They say it’s for security and performance.0:54 But critics say it neuters the most effective tools we have for blocking ads and trackers.1:00 And that leaves one major holdout: Firefox.1:02 On August tenth, they posted on Bluesky, quote: “Our support for uBlock Origin isn't going anywhere.” And for a lot of people, that’s it.1:11 That’s the whole ballgame.1:13 As one user put it, “This is officially Firefox's killer app.1:17 I will never go back to Chrome so long as uBlock Origin is supported on Firefox.” Now, while we're fighting for control over what we see online, a prominent cryptographer is warning about a different kind of control problem.1:31 Matthew Green, in a blog post from August fourteenth titled “Everything is about to go dark,” argues that the combination of powerful encryption and AI is about to make law enforcement surveillance almost impossible.1:45 He’s not talking about the future.1:47 He’s talking about right now.1:49 He says, quote: “I’m concerned that U.S.1:51 intelligence and law enforcement agencies are about to go dark, meaning that they’re going to suddenly lose a huge portion of their capability.” This isn't a new fear, but he argues AI-driven security is the accelerant that finally makes it a reality.2:07 And right on cue, as if to answer that very problem, Google announced a project called HEIR.2:13 It’s an open-source compiler that enables what they call private AI.2:18 The core technology is something called homomorphic encryption.2:22 This is the part that sounds like science fiction.2:25 It lets an AI model perform calculations on your data while it's still encrypted.2:30 The AI can do its job without ever actually seeing your private information.2:35 Think about healthcare or finance, where you want AI-powered features but can't risk exposing the raw data.2:42 Google's security blog says HEIR, quote, “unlocks cryptographically-secure private AI inference.” It’s a vision of a world where you get the benefits of AI without sacrificing privacy.2:53 And that same theme of control and openness is playing out in the Linux world, too.2:59 A project called RustDesk just released a preview build that solves a huge headache for a lot of people.3:05 It enables true unattended remote access on Wayland, which is the newer, more secure display system for Linux.3:12 This means you can remotely access your Linux machine, even from the login screen after a reboot, without anyone needing to be there to click "approve." For anyone managing servers or just helping family with their computers, this is a big deal.3:28 Other tools like AnyDesk and TeamViewer have struggled with this.3:32 RustDesk is getting it done, and it’s open source.3:35 Finally, a quick one that’s less about grand philosophy and more about practical control.3:41 Anthropic, the company behind the AI model Claude, published a guide on how to use their coding assistant, Claude Code, more efficiently.3:50 It’s full of tips on how to reduce token usage, which means reducing your cost.3:55 Things like clearing the conversation between tasks or setting the model’s effort level upfront.4:01 It’s a reminder that even as these AI tools get more powerful, learning how to pilot them effectively is a skill in itself.4:09 As the post says, “Being efficient with tokens doesn't mean using fewer of them overall.4:14 It means making sure the ones you do use go towards the thing you actually asked for.” It's about precision.4:21 It's about control.4:22 So let's put two of these threads together, because they represent two sides of the same, very important coin.4:29 On one side, you have Matthew Green’s warning: “Everything is about to go dark.” On the other, you have Google’s solution: HEIR, the compiler for private AI.4:39 This isn't just a technical debate.4:42 This is the central tension of our connected future.4:45 How do we build a secure, private world without creating a world that's totally opaque to legitimate investigation?4:52 Let's start with Green's argument, because to understand it, you have to see where we've been.4:58 This is a classic pattern.5:00 He traces the history of surveillance.5:02 Think back to 2002.5:04 Law enforcement wants to listen to a suspect's calls.5:07 What do they do?5:08 They get a warrant, go to the phone company, and they put a wiretap on a physical line connected to a payphone.5:15 The data—the conversation—is flowing in the clear, and they can intercept it.5:20 Fast forward to 2016.5:21 Now, almost everyone has a smartphone.5:24 Apps like Signal and WhatsApp have end-to-end encryption by default.5:28 The conversation is scrambled from the moment it leaves your phone to the moment it reaches the other person's.5:35 There's no central line to tap.5:37 The data is dark.5:38 The FBI saw this coming.5:40 Back in 2014, they launched a whole public relations campaign around this exact problem, calling it the "Going Dark" initiative.5:48 Their argument was that widespread, user-controlled encryption was making their jobs impossible.5:54 The tech community largely pushed back, saying that you can't build a "backdoor" for just the good guys.6:01 A vulnerability is a vulnerability, and bad actors will exploit it.6:05 And for a while, that was the stalemate.6:08 What Green is saying now is that AI is about to break that stalemate.6:12 Not by creating backdoors, but by making the front door so impossibly strong that nobody can get in.6:18 AI can be used to find and patch security holes faster than humans ever could.6:23 It can design more complex and robust encryption protocols.6:27 The result?6:28 The digital world gets locked down, hard.6:30 And the capabilities of law enforcement to investigate crime—from terrorism to child exploitation—could evaporate.6:38 It's a genuinely difficult problem.6:40 So what's the other side of the coin?6:42 What's the industry's answer?6:44 It’s projects like Google’s HEIR.6:46 Here’s the pattern twin for this.6:48 For years, the deal with tech was: you give us your data, and we give you a free service.6:54 Your search history, your photos, your location—it all gets fed into the machine to train models and sell ads.7:01 It's a Faustian bargain we all kind of accepted.7:04 Now, the pushback is immense.7:06 People want the smart features, but they're not willing to hand over the keys to their entire digital life anymore.7:13 Enter homomorphic encryption.7:15 Let’s be clear, the concept isn't new.7:18 But making it work in the real world, at scale, has been the challenge.7:22 Here's the simplest way to think about it.7:25 Imagine you want a jeweler to work on a diamond, but you don't trust them not to swap it for a fake.7:31 So you put the diamond in a locked, transparent box.7:34 The jeweler can put their tools into the box through special gloves, cut and polish the diamond, and hand the box back to you.7:43 They did the work, but they never actually touched the diamond itself.7:47 You, and only you, have the key to open the box and get the finished gem.7:52 That's what HEIR aims to do for AI.7:54 Your data is the diamond.7:55 It gets put in an encrypted "box." The AI model can perform its inference—it can analyze the photo, translate the text, check the medical scan—without ever decrypting the data.8:07 It does the work on the locked box.8:09 The result comes back to you, still encrypted.8:12 Google gets to run its service, you get the AI-powered feature, and your raw data is never exposed.8:18 It sounds perfect, right?8:20 A world where privacy and utility can coexist.8:23 So does this solve the "Going Dark" problem?8:25 No.8:26 Not really.8:26 And here's where the analogy breaks down, and it's the most important part.8:31 HEIR is designed for a specific scenario: a user sending their data to a company's server for processing.8:38 It protects your data from the company.8:40 It does NOTHING to solve the problem Matthew Green is talking about, which is two users communicating with each other using strong end-to-end encryption that the company—and by extension, law enforcement—can't read in the first place.8:55 HEIR is a tool for corporate data handling.8:58 It's not a backdoor for government surveillance.9:01 And there's the other catch: homomorphic encryption is still EXTREMELY slow and computationally expensive.9:08 It's like asking that jeweler to work with thick, clumsy oven mitts instead of surgical gloves.9:14 Yes, it's possible, but it's not practical for everything...9:18 yet.9:18 Google making an open-source compiler is a huge step toward making it more practical, but Green's problem is a five-alarm fire happening today.9:27 HEIR is a promising blueprint for a fireproof building we might be able to construct tomorrow.9:33 So what does it all add up to?9:35 You've got this incredible, grassroots, open-source energy pushing for more user control.9:41 You see it with Firefox and uBlock Origin.9:43 You see it with RustDesk on Linux.9:46 And then you have the big tech platforms like Google trying to solve the privacy problem on their own terms, with incredibly complex solutions like HEIR.9:55 And hovering over all of it is the government, watching its traditional sources of intelligence dry up, wondering what comes next.10:04 This week wasn't about a single breakthrough.10:06 It was about seeing the battle lines being drawn more clearly than ever.10:11 The fight for control over data is accelerating.10:14 On one side are tools that give you, the user, absolute power to lock your data down.10:20 On the other are the consequences of what happens when everyone does that all at once.10:25 The tools to lock down our digital lives are getting better, faster.10:29 The debate is no longer about if we can go dark, but what kind of society we'll have when we do.
