Government Gates and Quantum Leaps I 3rd June

President Trump's AI executive order sparks a heated debate about government oversight versus innovation freedom. Meanwhile, Microsoft drops bombshells at Build 2026 with quantum computing breakthroughs and a new AI assistant called Scout. Plus, OpenAI targets white-collar workers with specialized tools, and Google fights back against deepfake phone scams. When should the government step in to regulate AI, and when should it step back? A conversation that gets surprisingly heated.

Transcript

0:00

I keep going back and forth on this. I think I actually land on the side that this is a good thing. Really? Because I read the same story and I came out the other end deeply uncomfortable. But think about it. We're talking about AI models that could potentially be used for everything from cyber attacks to creating biological weapons. Shouldn't there be some kind of review process? Sure. But who gets to decide what's dangerous and what happens when that review process becomes a bottleneck that kills innovation? We're about to find out what happens when politics meets artificial intelligence. And the fact that industry pushback already forced them to walk it back to voluntary reviews, that tells you everything about how this is going to play out. You're listening to Build by AI. I'm Alex Shannon, and we just dropped you into what might be the most important conversation about AI regulation we've had yet. And I'm Sam Hinton. Look, today we're covering Trump's AI executive order that's got everyone talking, Microsoft's massive announcements at Build 2026, and some pretty wild developments in quantum computing. Plus, open AI is going off to your job specifically. They've got new tools for white-collar workers that are honestly pretty impressive. And Google's fighting deep fake phone scams, which, let me tell you, is becoming a bigger problem than most people realize. Alright, let's dive in. President Trump signed an executive order on AI oversight. But here's the thing. The original plan was mandatory government review of AI models before companies could release them. That would have been huge, but after industry pushback, what we actually got was voluntary pre-release reviews. So companies can choose to submit their models for government review, but they don't have to. It's basically asking nicely instead of requiring it. And that's exactly why I'm uncomfortable with this whole thing. It's not that I'm against the idea of reviewing dangerous AI models. I'm against the government being the arbiter of what gets released and what doesn't. But Sam, come on. We're talking about models that could potentially be used to design bioweapons, plan cyber attacks, or manipulate elections on a massive scale. Shouldn't there be some guardrails? Sure. But think about the precedent this sets. Today it's voluntary, but what happens next time? What happens when the review process becomes politicized? What happens when it takes six months to get approval and startups can't compete because they can't afford to wait? That's a fair point. And honestly, the fact that industry pressure was enough to make them walk it back to voluntary shows you how this is going to go. If companies don't like the voluntary system, they'll just ignore it. Exactly. And and here's what really bothers me. Who's doing these reviews? You know, are we talking about technical experts who understand AI, or are we talking about government bureaucrats who might not even know what a neural network is? The order doesn't really specify, which is another problem. But I keep coming back to this. If we don't figure out some kind of oversight system now, while AI is still developing, we might not get another chance. I hear you, but I think the better approach is industry self-regulation combined with existing laws. We don't need new bureaucracy, like we need companies to take responsibility, and we need to enforce the laws we already have. But how's that working out so far? I mean, we're already seeing AI being used for fraud, disinformation campaigns, harassment. The self-regulation approach feels like it's not keeping pace with the technology. Okay. You've got a point there. But here's my concern. Once you create a government review process, it's almost impossible to scale it back. What starts as voluntary becomes mandatory. What starts as security focused expands to content moderation. That's a slippery slope argument, though. We regulate pharmaceuticals, nuclear technology, aircraft design, lots of things that could be dangerous. Why should AI be different? Aaron Ross Powell Because AI is fundamentally different from those things. A drug does one thing, a plane does one thing. But I models are general purpose tools that can be used for thousands of different applications. You can't review every possible use case. Maybe, but you can review the capabilities of the model itself. Like, does this model make it easier to create biological weapons? Can it be used to generate convincing deep fakes? Those are answerable questions. Sure. But then what? Let's say a model can generate convincing text. Do you ban it because someone might use it for fraud? Do you restrict it because someone might use it for disinformation? Where do you draw the line? I don't know, but I'm not convinced that no line is the right answer either. And maybe the voluntary nature of this might be the worst of both worlds. Not strong enough to actually prevent dangerous releases, but bureaucratic enough to slow down legitimate research. My prediction is that the big tech companies will participate because they can afford to, and smaller players will just ignore it entirely. So you end up advantaging the incumbents. Which brings up another issue. The speed of the industry pushback. How quickly they were able to get this walked back from mandatory to voluntary tells you a lot about who has influence in these conversations. Right, and that's both encouraging and concerning. Encouraging because it shows the industry can push back against overreach. Concerning because it raises questions about whether any meaningful regulation is actually possible. The timing is interesting too. This comes right as AI capabilities are advancing really rapidly. In six months, we might be dealing with models that are significantly more powerful than anything we have today. Exactly, and that's why I think the focus should be on building good governance structures and technical standards now, rather than creating review processes that might not be able to keep up with the pace of development. Alright, I think we're going to have to agree to disagree on this one. But it'll be really interesting to see how this plays out in practice. Whether companies actually use the voluntary review process and what happens the first time something goes wrong. Alright, let's shift gears to something a bit more exciting. Microsoft just announced their Majorana 2 quantum chip, and they're claiming it significantly cuts the timeline to achieving useful quantum computing. This follows their previous breakthrough with the Majorana 1 processor, but apparently Majorana 2 is a much bigger leap forward. The question is, what does significantly cut the timeline actually mean? Yeah, uh that's the key question, because we've been hearing quantum computing is five years away for like 15 years now. But Microsoft's approach with these Majorana chips is actually pretty different from what IBM and Google are doing. Can you explain that for people who aren't deep in the quantum weeds? What makes Microsoft's approach special? So most quantum computers today are incredibly fragile. They need to be kept colder than outer space, and any tiny vibration can mess them up. Microsoft is betting on something called topological qubits, which should be much more stable. Think of it like the difference between balancing a pencil on its tip versus laying it flat on a table. Current quantum computers are like the pencil on its tip. Really hard to keep stable. Um Microsoft's approach is more like the pencil lying flat. That's a great analogy. But here's what I'm curious about. If this approach is so much better, why isn't everyone doing it? What's the catch? The catch is that it's been really, really hard to actually build these topological qubits. Um Microsoft has been working on this for over a decade, and they've had some false starts. But if Majorana 2 actually works as advertised, it could be game changing. Right, and that's the thing with quantum computing announcements. There's always this question of whether it's a real breakthrough or just incremental progress that's being oversold. How do we tell the difference? Good question. The key metrics to watch are error rates and coherence times. Basically, how often do the qubits make mistakes and how long can they maintain their quantum state? If Majorana 2 shows significant improvements in both, that's a big deal. And when we're talking about useful quantum computing, what are we actually talking about? What's the first real-world application we're likely to see? Probably drug discovery or material science. Quantum computers are really good at simulating molecular interactions, which is incredibly hard for classical computers. Imagine being able to design new medications or batteries in a fraction of the time it takes now. That could be huge for healthcare and climate change. But I'm also thinking about the darker applications. Quantum computers could potentially break a lot of current encryption methods, right? Absolutely. And that's why there's this race to develop quantum resistant encryption. The good news is that we're probably still years away from quantum computers that can break current encryption, but it's definitely something to keep on your radar. Years away based on the old timeline? Or years away based on Microsoft's new timeline? Because if they're right about significantly reducing the time to useful quantum computing, maybe we need to be thinking about this more urgently. The quantum computers that can break encryption need to be much more powerful than the ones that can help with drug discovery. We're talking about different orders of magnitude. So the beneficial applications might come first, which is good. But it raises this interesting question about how you manage a technology that has both incredible positive potential and serious security implications. Right. And unlike AI, where you know we're sort of figuring out the implications as we go. With quantum computing, we can see some of the risks coming. You know, that gives us time to prepare, but it also means we need to start taking those preparations seriously now. So with Microsoft's announcement, are we talking about moving from five years away to two years away? Or is this more incremental progress? Hard to say without more technical details, but the fact that they're calling it a significant reduction in timeline suggests this is more than incremental. Um I'd love to see some actual benchmarks and comparisons to their previous work. The competitive dynamics here are interesting too. If Microsoft really has cracked the stability problem with topological qubits, that could give them a huge advantage over Google and IBM's approaches. Absolutely. And quantum computing is one of those winner-take most technologies. If your approach is fundamentally more stable and scalable, you don't just win by a little bit, you win by a lot. Although, let's be honest, Microsoft has been promising breakthroughs in topological quantum computing for a while now. How confident should we be that this time is different? That's fair skepticism. But the progression from Majorana 1 to Majorana 2 suggests they're making real progress, not just incremental improvements. The proof will be in independent verification and actual performance benchmarks. Speaking of Microsoft announcements from Build 2026, they also launched something called Scout, which is a new AI personal assistant that's inspired by something called OpenClaw. Now I'll be honest, I'm not familiar with OpenClaw, but Scout is designed to integrate with Microsoft 365. So we're talking about an AI assistant that works across Word, Excel, Outlook, all those tools. All. Oh man. OpenClaw is really interesting. It's been more of an open source project that's focused on really sophisticated task automation, like way beyond what Siri or Alexa can do. Think more like having a digital employee than a voice assistant. That sounds both incredibly useful and mildly terrifying. What kind of tasks are we talking about here? Well, with the Microsoft 365 integration, imagine telling Scout, prepare a quarterly report on our sales performance. And it goes out, pulls data from multiple spreadsheets, creates charts, writes the narrative, and formats everything in PowerPoint. Or you could say schedule a meeting with everyone who worked on the Johnson project and send them the latest project files. And it just handles all of that coordination automatically. Okay, that's actually really impressive. But here's my question. How good is it going to be? Because we've seen a lot of AI assistants that are great for demos but frustrating to use in real life. Yeah, that's the million-dollar question. Microsoft has a huge advantage here because they own the entire stack, the AI models, the office software, the cloud infrastructure. That integration could make Scout way more reliable than assistants that have to work across different platforms. But it also means you're locked into Microsoft's ecosystem. If you're using Google Workspace or other tools, Scout probably isn't going to be as useful. Exactly. And that's definitely intentional. Microsoft is betting that Scout will be compelling enough to get people to switch to Microsoft 365 or stick with it instead of moving to competitors. The timing is interesting too, because this comes right as we're seeing a lot of companies trying to figure out how to actually use AI productively. A lot of businesses are still struggling to get value out of ChatGPT or other general AI tools. Right. And Scout could be the answer to that problem. Instead of employees having to figure out how to prompt ChatGPT to help with their Excel spreadsheet, they just tell Scout what they need in plain English. The OpenClaw inspiration is really smart too. OpenClaw has been developing some of the most sophisticated automation capabilities, but it's been more of a technical user tool. Microsoft taking those ideas and making them accessible to regular business users could be huge. And think about the competitive implications. If Scout works as advertised, it could give Microsoft a significant advantage in the business productivity space. Google and others are going to have to respond with their own equivalent tools. But there are also some concerning aspects here. If Scout can automatically read through all your emails, access all your files, and coordinate with other people on your behalf, that's a lot of access and control to give an AI system. Absolutely. And what happens when Scout makes a mistake? If it sends the wrong document to the wrong person or schedules a meeting at the wrong time, who's responsible for that? That's going to be a learning process for everyone. I I suspect we'll see Scout start with lower stakes tasks and gradually expand its capabilities as users get more comfortable with it. I'm curious to see the pricing on this. Is it going to be included with Microsoft 365 subscriptions, or is this going to be another premium add-on? Microsoft didn't announce pricing yet, but my guess is it'll be a premium feature, at least initially. They'll probably use it to drive people to higher tier subscriptions. Keep an eye on the rollout. I bet they start with enterprise customers before bringing it to consumers. The enterprise focus makes sense. Business users probably have more predictable workflows that Scout can automate effectively, and they're more likely to pay premium prices for productivity gains. Plus, enterprise customers have IT departments that can handle the setup and security configurations. Rolling this out to hundreds of millions of consumer users right away would probably be a support nightmare. And speaking of AI tools coming for white-collar jobs, OpenAI just released six new codex plugins that are designed for specific professions. We're talking about data analytics, creative production, sales, product design, and investment roles. Each tool apparently bundles integrations and job-specific context. So instead of using general chat GPT, you'd use a version that's specifically designed for your type of work. This is huge, and honestly, it's what I've been waiting for. Generic AI assistants are fine, but they don't really understand the specific workflows and tools that different professionals use every day. Give me an example. What would the difference be between regular ChatGPT and, say, the Data Analytics Codex plugin? So the data analytics version would probably have direct integrations with tools like Tableau, Python libraries, SQL databases. It would understand common data science workflows and be able to write code that actually works in those specific environments. Instead of having to copy and paste code and fix all the errors, you could say create a visualization showing sales trends by region. And it would pull the data, clean it, and generate the chart in whatever tool you're using. That's a pretty compelling value proposition. But I'm wondering about the quality. How good can these specialized tools really be compared to having a human expert? Well, they're probably not going to replace senior analysts or designers anytime soon, but they could be incredibly powerful for junior employees or for handling routine tasks that senior people don't want to spend time on. The list of professions they're targeting is really interesting. Data analytics makes sense. Sales makes sense. But creative production, that's got to have creative professionals pretty worried about job security. Yeah, but I think the key word here is production rather than creation. My guess is this this tool is more about helping creative professionals work faster, automating the tedious parts of production so they can focus on the actual creative work. That makes sense. Like having an AI that can handle color correction, file formatting, asset management, that kind of thing. Exactly. You know, the tools are available within the Codex app, so this isn't just about ChatGPT getting better. OpenAI is building a whole ecosystem of specialized AI tools. That's a pretty smart business strategy. The investment banking and equity investing tools are particularly interesting to me. Those are fields where small mistakes can cost millions of dollars. Are we really ready for AI to be handling that kind of work? I think it depends on how they're used. If it's helping analysts run financial models or generate first drafts of research reports, that could be really valuable. But if it's making actual investment decisions, that's a different story entirely. Right. And the liability questions are huge. If an AI-generated financial model has an error that leads to bad investment decisions, who's responsible? The user, open AI, the investment firm? Those legal frameworks are still being figured out, but you know, I suspect these tools will start with more support and analysis functions rather than decision-making functions, at least initially. The competitive dynamics here are interesting too. If these codex tools are really good, they could give companies using them a significant advantage over competitors who are still doing everything manually. Absolutely. We might see a situation where using AI tools becomes necessary just to keep up, not to get ahead. That could accelerate adoption really quickly across these white-collar professions. But that also raises questions about the digital divide. Companies and professionals who can afford these premium AI tools could pull further ahead of those who can't. That's a good point. OpenAI hasn't announced pricing for these codex plugins yet, but they're probably not going to be free. And if they require integrations with expensive enterprise software that could limit access even further. The product design tool could be really interesting for smaller companies. Instead of hiring expensive design consultants, you might be able to get AI assistance with product development and design decisions. True. Probably not replace, but maybe augment. Like helping designers explore more options quickly, or identifying potential problems with designs before they get built. And that's probably the right way to think about all these tools, not as replacements for human professionals, but as ways to make human professionals more capable and efficient. Microsoft Build 2026 was apparently packed with announcements beyond just Scout and the quantum chip. They also announce new surface hardware. Yeah, and build is usually where Microsoft shows off their vision for the next year of development tools and platforms. The fact that they're putting so much focus on AI integration across their entire product lineup tells you where they think the industry is heading. AI into their applications in useful ways. Right. And Microsoft is in a really good position here because they have Azure for the cloud infrastructure, they have the partnership with OpenAI for the AI models, and they have all the development tools. They're basically offering a one-stop shop for EEI development. Satya Nadella was clearly making the case that Microsoft is going to be the AI platform for developers and businesses. Between the quantum computing announcements, Scout and the new surface hardware, they're covering the whole stack. And that vertical integration strategy could be really powerful if developers buy into it. But it also creates dependency on Microsoft's ecosystem in a way that some companies might be uncomfortable with. The always-on personal assistant aspect is particularly interesting. That suggests they're thinking about AI as something that's constantly running in the background, not just something you invoke when you need it. Which has huge implications for privacy and battery life and user experience. An always-on AI assistant could be incredibly helpful, but it also means your device is constantly listening and processing. That's going to require a lot of trust. Scaling their Clawed Mythos model to critical infrastructure in over 15 countries. We're talking about power, water, healthcare, communications. If confirmed, that's actually pretty significant. Critical infrastructure security is one of those areas where AEI could be incredibly valuable, helping detect threats, optimize operations, prevent failures before they happen. But it also makes me nervous. If Claude Mythos is protecting critical infrastructure and something goes wrong, the consequences could be pretty severe. True, but the alternative might be worse. A lot of critical infrastructure is running on decades-old systems that are vulnerable to increasingly sophisticated cyber attacks. A AI might be our best defense against AI-powered attacks. The scale is interesting. 150 organizations across 15 countries. That suggests this isn't just a pilot program, but a pretty significant deployment of AI in security critical systems. And the fact that it's happening through Project Glasswing, which is Anthropic's security vulnerability program, suggests they're taking the safety aspect seriously. But I'd love to see more details about what safeguards and oversight mechanisms are in place. So we should be cautious about drawing too many conclusions. But if true, it represents a major step toward AI systems managing critical infrastructure at scale. Right. And that's exactly the kind of application where the stakes are so high that you really want multiple independent sources confirming what's happening. The implications for national security and public safety are huge. Speaking of AI-powered attacks, early reports suggest Google is rolling out fake call detection technology to combat AI deep fake impersonation scams. Apparently, scammers are using AI to mimic authority figures and family members. It's basically an AI arms race. A AI-generated voices getting better, but AI detection getting better too. The fact that scammers are spoofing trusted numbers makes this even more dangerous. You get a call from what looks like your bank's number. And it's an AI voice that sounds exactly like a real customer service representative. And the scary thing is how quickly this technology has become accessible. A few years ago, creating convincing voice clones required expensive equipment and technical expertise. Now you can do it with a smartphone app and a few audio samples. Google's response is interesting because it's reactive rather than proactive. They're building defenses against a problem that's already happening rather than trying to prevent the technology from being misused in the first place. Which might be the right approach, honestly. You probably can't prevent deep fake technology from existing. So building better detection systems might be more effective than trying to control the technology itself. Right, that was the original plan. But industry pushback was swift and effective. The fact that they they walked it back so quickly tells you a lot about the political dynamics around AI regulation right now. It makes me wonder if we're going to see this pattern repeat. Initial proposals for strict regulation, followed by industry lobbying, followed by watered down voluntary measures. Probably, at least until something goes really wrong and forces stronger action. The question is whether we'll be proactive about regulation or reactive. The contrast between the two versions is striking. Mandatory reviews would have been a major shift in how AI development works. Voluntary reviews are more like a suggestion that companies can ignore, Carl. And it shows how much influence the tech industry has in these policy discussions. When major companies push back against proposed regulations, they often get walked back pretty quickly. Which raises questions about whether meaningful AI regulation is actually possible in the current political environment, or whether the industry will always be able to water down anything that might slow them down. That's the tension we're going to be dealing with for the foreseeable future. How to balance innovation and safety when the innovators have a lot of political and economic power. If you zoom out and look at everything we covered today, there's this interesting tension between innovation and control. Microsoft is pushing the boundaries with quantum computing and AI assistance. OpenAI is building specialized tools for different professions. But at the same time, we're seeing the first real attempts at government oversight, companies building defensive technologies against AI-powered scams, and critical infrastructure being protected by AI systems. It feels like we're at this inflection point where AI is becoming powerful enough that we have to take both its potential and its risks seriously. The question is whether we can figure out the right balance. We're not just talking about chatbots anymore. We're talking about AI assistants that can do complex work tasks and quantum computers that might actually be practical soon. The thing that strikes me is how much of this is happening simultaneously. Government trying to regulate AI, companies building AI defenses against AI attacks, breakthrough hardware, specialized professional tools, it's all accelerating at once, is Hermann Doch. Right, and I think that's why these conversations about regulation and oversight are so important. We need to figure out the rules while there's still time to shape how this technology develops, not just react to whatever happens. But the speed of change makes that really difficult. By the time you've developed regulations for today's AI capabilities, the technology has moved on to something completely different. Which is why I think the focus needs to be on principles and frameworks rather than specific technical requirements. You need governance structures that can adapt as the technology evolves. The competition aspect is huge too. Microsoft is making a play to dominate multiple layers of the AI stack, from quantum hardware to productivity software to development platforms. That kind of vertical integration could give them enormous power. And that's happening across the industry. Google, Amazon, OpenAI, Anthropic, everyone is trying to build comprehensive AI ecosystems rather than just point solutions. The winners could end up controlling huge parts of the economy. Which brings us back to the regulation question. If a few companies end up controlling most AI infrastructure, that creates concentration of power that might need to be addressed through antitrust or other policies. But at the same time, we're seeing AI being used to solve real problems, protecting critical infrastructure, helping with professional work, defending against scams. The technology isn't inherently good or bad. Exactly. It's about how it's used and who controls it. The same AI technology that can help a data analyst work more efficiently can also be used to create convincing scams. The same quantum computers that can help design new medicines could also break encryption. You can't just regulate the technology itself. You have to think about use cases, access, concentration of power, international competition, all of it. The international dimension is really important too. If the US regulates AI development too strictly, does that just push innovation to other countries? But if we don't regulate it enough, do we end up with unsafe or misused technology? And meanwhile, the technology keeps advancing. Today's announcements about quantum computing and specialized AI tools show that we're moving toward a world where AI is deeply integrated into critical systems and professional workflows. Which means the window for shaping how this develops is probably shorter than we think. Once AI systems are managing critical infrastructure and handling sensitive professional tasks at scale, it becomes much harder to change course. That's why I think both sides of the regulation debate we had earlier have valid points. We need oversight, but we also need to be careful not to kill innovation or create bureaucratic bottlenecks that just advantage incumbents. Alright, that's a wrap for today. As always, if you're getting value out of these conversations, the best way to support the show is to subscribe and share it with someone who should be keeping up with AI news. And honestly, with everything happening in AI right now, daily updates feel more essential than ever. Tomorrow we'll be back with whatever the AI world throws at us next. Until then, I'm Alex Shannon. And I'm Sam Hitten. See you tomorrow on Build by AI.