Amazon's $5B Anthropic Bet and the NSA's Secret AI

Alex Shannon: Picture this: It’s early 2027, and you’re sitting in a coffee shop in downtown Seattle. At the table next to you, an Amazon executive is explaining to their friend how they just made five billion dollars appear out of thin air by simply promising to spend twenty times that amount on their own company’s services.

Sam Hinton: Meanwhile, across the country in Fort Meade, NSA analysts are running queries through an AI system that their own Pentagon colleagues have been explicitly told not to touch.

Alex Shannon: And somewhere in between, a security researcher just discovered that the protocol connecting all these AI systems together has a vulnerability that could bring down the entire supply chain.

Sam Hinton: Because of something that just happened this week. Actually, several somethings.

Alex Shannon: The AI industry just had one of its weirdest weeks yet, and we’re about to break it all down.

Alex Shannon: You’re listening to Build By AI, I’m Alex Shannon, and what we just described isn’t science fiction - it’s Monday morning in the AI world.

Sam Hinton: And I’m Sam Hinton. Today we’re diving into Amazon’s mind-bending circular investment strategy with Anthropic, why the NSA is using AI that the Pentagon apparently doesn’t want them to have, and a security vulnerability that should have every AI company checking their locks twice.

Alex Shannon: Plus, we’ve got some wild stats on AI-generated music, Florida’s attempt at an AI Bill of Rights, and a nuclear AI startup that just lost its entire executive team.

Sam Hinton: It’s Monday, April 21st, 2026, and honestly, I’m still trying to wrap my head around some of these stories.

Alex Shannon: Alright, let’s start with the big one.

Anthropic takes $5B from Amazon and pledges $100B in cloud spending in return

Alex Shannon: So Amazon just invested five billion dollars in Anthropic, which sounds like normal venture capital news until you hear the other half of the deal. Anthropic has committed to spend one hundred billion dollars on AWS cloud services in return.

Sam Hinton: Wait, let me make sure I have this right. Amazon gives Anthropic five billion dollars, and then Anthropic promises to give Amazon one hundred billion dollars over time by buying their cloud services?

Alex Shannon: That’s exactly right. It’s what they’re calling a circular investment deal. Amazon is essentially paying Anthropic to become their biggest cloud customer.

Sam Hinton: Dude, this is like me giving you fifty bucks and then you promising to spend a thousand dollars at my lemonade stand. Except we’re talking about billions and the future of AI.

Alex Shannon: But here’s what I’m trying to understand - is this actually good business for Amazon, or is this some kind of accounting gymnastics to lock in cloud revenue?

Sam Hinton: It’s brilliant business, actually. Amazon gets to book five billion in investment losses but lock in twenty times that in guaranteed revenue over the next several years. Plus, they get a say in how one of the top three AI companies develops.

Alex Shannon: OK but let’s play devil’s advocate here. Doesn’t this create some weird incentives? Anthropic now has to spend a hundred billion on AWS whether that’s the best choice for their technology or not.

Sam Hinton: That’s the concerning part. We’re basically watching Amazon use their cloud dominance to buy influence in the AI space. It’s like if Ford invested in a taxi company but only if they promised to buy Ford cars for their entire fleet.

Alex Shannon: And the reporting says this is another circular AI deal for Amazon, which suggests they’ve done this before.

Sam Hinton: Right, and that’s the pattern we should be watching. Amazon isn’t just competing in the AI space - they’re using their cloud business to essentially create financial dependencies with AI companies.

Alex Shannon: What does this mean for smaller AI startups who can’t make hundred billion dollar commitments?

Sam Hinton: They get squeezed out of the premium partnership deals. Amazon gets to pick the AI winners and losers not just through technical capabilities, but through financial engineering.

Alex Shannon: That raises a question about market competition though. If Amazon can essentially buy their way into controlling AI development through these circular deals, are we looking at an antitrust issue?

Sam Hinton: I think we might be. When your cloud infrastructure business becomes so dominant that you can use it to influence which AI companies succeed, that sounds like the kind of vertical integration that regulators usually don’t like.

Alex Shannon: But from Anthropic’s perspective, I can see why this makes sense. They get five billion in immediate capital and access to Amazon’s massive infrastructure without having to build it themselves.

Sam Hinton: True, but it also means they’re locked into AWS pricing and capabilities for what could be decades. If Google or Microsoft develops better AI infrastructure, Anthropic can’t just switch.

Alex Shannon: And that hundred billion commitment - let’s break that down. That’s not just compute costs, that’s also data storage, networking, all the ancillary services that AWS provides. Amazon is essentially guaranteeing themselves a massive recurring revenue stream.

Sam Hinton: Right, and they probably structured it so that if Anthropic doesn’t hit those spending targets, there are penalties or Amazon gets more equity. These deals are never as simple as they appear on the surface.

Alex Shannon: The timing is interesting too. This comes right as we’re seeing massive AI compute demands and questions about whether startups can afford to scale their models. Amazon is basically saying ‘we’ll solve your funding problem if you become our exclusive customer.’

Sam Hinton: And for listeners who are developers or working at smaller AI companies, this should be a wake-up call. The big cloud providers aren’t just infrastructure vendors anymore - they’re actively picking which AI companies they want to succeed.

Alex Shannon: What’s your prediction on how this plays out? Do we see Microsoft doing similar deals with OpenAI, Google with their AI partners?

Sam Hinton: I think we’re about to see a whole wave of these circular investment deals. If it works for Amazon and Anthropic, why wouldn’t every major cloud provider start using the same playbook?

Alex Shannon: Keep an eye on this because if it works, we’re probably going to see Microsoft, Google, and others start copying this playbook with their own AI investments.

Sam Hinton: And honestly, if you’re an AI startup that can’t make these kinds of massive infrastructure commitments, you might want to start thinking about your competitive position. Because the barriers to entry just got a lot higher.

NSA spies are reportedly using Anthropic’s Mythos, despite Pentagon feud

Alex Shannon: Now this next story is fascinating and a little concerning. The NSA is reportedly using Anthropic’s restricted AI model called Mythos, despite existing tensions with the Pentagon over Anthropic.

Sam Hinton: Wait, hold on. The NSA and the Pentagon are part of the same government, right? How does one agency end up using AI that another agency is apparently feuding with the company about?

Alex Shannon: That’s exactly what makes this so weird. Mythos is described as a restricted model, which suggests it’s not something you can just sign up for on Anthropic’s website.

Sam Hinton: This screams of bureaucratic dysfunction to me. The left hand doesn’t know what the right hand is doing, except the hands are different intelligence agencies and they’re all playing with the same AI company.

Alex Shannon: But let’s think about what Mythos might be. If it’s restricted and the NSA is using it, this could be a specialized intelligence-gathering AI that’s designed specifically for surveillance or analysis work.

Sam Hinton: That’s actually terrifying when you think about it. We’re talking about the most sophisticated AI models being deployed by spy agencies, potentially without proper oversight if there’s conflict between agencies.

Alex Shannon: The Pentagon feud angle is really intriguing though. What could the Pentagon’s issue with Anthropic be that the NSA is apparently ignoring?

Sam Hinton: My guess? Either security concerns, ethical guidelines, or competition with Pentagon-backed AI projects. But the NSA clearly thinks whatever Mythos does is worth the inter-agency drama.

Alex Shannon: Let’s think about this from a practical standpoint. The NSA has very specific needs - they’re doing signals intelligence, communications analysis, pattern recognition on massive scales. A restricted AI model would give them capabilities that regular models can’t provide.

Sam Hinton: Right, and ‘restricted’ could mean it has access to classified training data, or it can process classified information, or it has capabilities that would be dangerous in civilian hands. The fact that it’s called Mythos is interesting too - that sounds like a code name.

Alex Shannon: But here’s what bothers me about this situation. If the Pentagon has legitimate concerns about Anthropic - whether it’s security, ethics, or competition - shouldn’t that be a government-wide policy?

Sam Hinton: You would think so, but intelligence agencies have always operated with more autonomy than other government departments. The NSA might be saying ‘we don’t care about Pentagon politics, this tool helps us do our job.’

Alex Shannon: This raises huge questions about AI governance in government. Should spy agencies be able to unilaterally decide which AI tools to deploy?

Sam Hinton: Absolutely not. This is exactly why we need clear federal AI policies. You can’t have one part of government saying ‘this AI is too risky’ while another part is already using it for national security.

Alex Shannon: And the timing is interesting - this comes out right after Amazon’s big Anthropic investment. Suddenly everyone wants a piece of Anthropic’s technology.

Sam Hinton: Which makes you wonder if the government use was part of what made Anthropic attractive to Amazon in the first place. Nothing says ‘proven technology’ like spy agencies betting their operations on it.

Alex Shannon: But think about the optics here. Amazon just committed to a hundred billion dollar relationship with a company that has secretive government contracts and internal Pentagon conflicts. That’s a lot of political risk.

Sam Hinton: Unless Amazon sees the government relationships as a feature, not a bug. Having the NSA as a customer is probably good for business, even if it comes with Pentagon drama.

Alex Shannon: The broader concern here is transparency. If government agencies are using advanced AI for intelligence work, shouldn’t there be some public awareness of the capabilities and limitations?

Sam Hinton: That’s the fundamental tension with intelligence agencies and AI. They want the most powerful tools available, but they also want to keep their capabilities secret. That makes oversight really difficult.

Alex Shannon: What worries me is that we might be creating a situation where AI development is being driven by intelligence community needs rather than civilian benefits. If spy agencies are the primary customers for advanced AI, that shapes how the technology develops.

Sam Hinton: And it creates incentives for AI companies to build capabilities that serve intelligence gathering rather than solving problems for regular people. The money talks, and if government agencies are writing big checks, that’s where the innovation goes.

Alex Shannon: For people listening who work in AI development or policy, this story should be a wake-up call. We need clear rules about government AI use before these ad hoc relationships become entrenched.

Anthropic MCP Design Vulnerability Enables RCE, Threatening AI Supply Chain - The Hacker News

Alex Shannon: Speaking of concerning developments with Anthropic, security researchers have discovered a design vulnerability in their MCP - that’s Model Context Protocol - that enables remote code execution attacks.

Sam Hinton: OK this is actually a huge deal. Remote code execution means an attacker can potentially run whatever code they want on systems using this protocol. That’s like finding out the lock on your front door can be opened by anyone with a paperclip.

Alex Shannon: And the reporting specifically mentions this threatens the broader AI supply chain, not just Anthropic’s systems. Help me understand what that means.

Sam Hinton: MCP is essentially how different AI systems talk to each other and share context. If there’s a fundamental design flaw in that communication protocol, it’s like having a vulnerability in email - it affects everyone using the system, not just one company.

Alex Shannon: But wait, didn’t we just talk about Amazon investing billions in Anthropic and the NSA using their restricted models? A supply chain vulnerability seems like terrible timing.

Sam Hinton: That’s exactly what makes this so concerning. You’ve got massive financial investments and government agencies betting on Anthropic’s technology right as we discover there might be fundamental security issues.

Alex Shannon: Let’s break down what remote code execution actually means for people who aren’t security experts. If someone exploits this vulnerability, what can they actually do?

Sam Hinton: They can potentially run any program they want on the target system. That could mean stealing data, installing malware, using the system to attack other systems, or even just deleting everything. It’s basically total compromise.

Alex Shannon: And because this is in a protocol that connects AI systems, an attacker could potentially jump from one system to another through the AI supply chain?

Sam Hinton: Exactly. If Company A uses MCP to connect to Company B’s AI service, and Company B gets compromised, the attacker might be able to pivot to Company A’s systems through that connection. It’s like a domino effect.

Alex Shannon: What I want to know is whether this is something that can be patched, or if it’s a deeper architectural problem with how they designed the protocol.

Sam Hinton: The fact that they’re calling it a ‘design vulnerability’ rather than just a bug suggests it might be the latter. Design vulnerabilities often require rebuilding things from scratch rather than just applying a patch.

Alex Shannon: That’s concerning because if it’s a fundamental design issue, fixing it could break compatibility with existing integrations. Companies might have to choose between security and functionality.

Sam Hinton: And given how fast the AI industry moves, I bet there are tons of companies that built integrations with MCP without doing thorough security reviews. They just trusted that Anthropic had done the security work.

Alex Shannon: And if other AI companies have built integrations with MCP, they’re all potentially vulnerable too.

Sam Hinton: Right, and this is why security experts have been warning about AI supply chain risks. When these models and protocols become critical infrastructure, vulnerabilities don’t just affect one company - they cascade through the entire ecosystem.

Alex Shannon: This also raises questions about the due diligence process for those big investments and government contracts. Was anyone actually auditing the security architecture?

Sam Hinton: Probably not deeply enough. The AI industry has been moving so fast that security reviews haven’t kept pace. Everyone’s been focused on capabilities rather than vulnerabilities.

Alex Shannon: But think about the implications here. The NSA is using Anthropic’s restricted models, Amazon just bet billions on the company, and now we find out there’s a fundamental security flaw that could affect the entire AI supply chain.

Sam Hinton: It’s like finding out that the foundation of a skyscraper is cracked after everyone’s already moved in. The question is whether this can be fixed without bringing the whole thing down.

Alex Shannon: And what’s particularly concerning is that this affects the AI supply chain. We’re not just talking about one company’s security - we’re talking about the interconnected web of AI services that companies rely on.

Sam Hinton: Exactly. Your company might have great security practices, but if you’re using an AI service that connects through MCP to other services, you’re only as secure as the weakest link in that chain.

Alex Shannon: For developers and companies using Anthropic’s tools, what should they be doing right now?

Sam Hinton: First, check if you’re using MCP in your integrations. Second, implement additional security layers that don’t rely on the protocol being secure. And third, have a backup plan in case this turns out to be unfixable quickly.

Alex Shannon: And honestly, this should be a wake-up call for the entire AI industry about security practices. We can’t just move fast and break things when those things are critical infrastructure that other companies depend on.

CEO and CFO suddenly depart AI nuclear power upstart Fermi

Alex Shannon: Now let’s shift gears to something completely different. Early reports suggest that both the CEO and CFO of an AI nuclear power startup called Fermi have suddenly departed the company.

Sam Hinton: Wait, AI nuclear power? That sounds like someone took two of the most complicated industries and decided to mash them together. What exactly does this company do?

Alex Shannon: From what we know, they operate an AI campus in Texas and were co-founded by former U.S. Energy Secretary Rick Perry. The idea seems to be using nuclear power to fuel AI data centers.

Sam Hinton: OK that actually makes sense from a power consumption standpoint. These big AI training runs need massive amounts of electricity, and nuclear is reliable baseload power. But losing both your CEO and CFO suddenly? That’s never a good sign.

Alex Shannon: The reporting mentions the company has faced headwinds, though it doesn’t specify what kind. Given that Rick Perry was involved, I’m wondering if this is more about regulatory challenges than technical ones.

Sam Hinton: Nuclear power and AI are both heavily regulated industries. Trying to combine them probably means dealing with the Department of Energy, the Nuclear Regulatory Commission, and whoever oversees AI development. That’s a bureaucratic nightmare.

Alex Shannon: But the timing of both executives leaving simultaneously suggests this wasn’t planned succession. This sounds like something went very wrong very quickly.

Sam Hinton: Yeah, you don’t lose your entire executive team unless there’s either a major strategic disagreement, funding issues, or regulatory problems that make the business model unviable.

Alex Shannon: Let’s think about the practical challenges here. Building nuclear facilities takes years of regulatory approval, massive upfront capital, and specialized expertise. That’s before you even get to the AI side of the equation.

Sam Hinton: And the liability issues must be insane. If something goes wrong with a traditional data center, you might lose some data or have downtime. If something goes wrong with a nuclear facility, you’re talking about evacuation zones and environmental cleanup.

Alex Shannon: Plus, AI workloads are notoriously unpredictable. Sometimes you need massive amounts of power for training runs, sometimes systems are idle. Nuclear power plants are designed for steady, consistent output.

Sam Hinton: That’s a really good point. Nuclear plants don’t scale up and down quickly like other power sources. If your AI workloads are variable, you’re either wasting nuclear capacity or you need backup power sources anyway.

Alex Shannon: What’s interesting is the broader trend here. We’ve seen multiple companies trying to solve AI’s energy problem - some with renewables, some with more efficient chips, and apparently some with nuclear power.

Sam Hinton: The energy costs of AI are becoming a real limiting factor. If you can’t solve the power equation, you can’t scale these systems. Nuclear actually makes theoretical sense, but the practical challenges are enormous.

Alex Shannon: And Rick Perry’s involvement is interesting. He was Energy Secretary under Trump, so he presumably knows the regulatory landscape. If even he couldn’t make this work, that says something about how difficult this space is.

Sam Hinton: It also makes you wonder about the timeline expectations. Maybe investors thought this would be a quick win - build some nuclear facilities, plug in AI systems, profit. But the reality of nuclear development is measured in decades, not years.

Alex Shannon: The fact that they had an AI campus in Texas suggests they were actually operational to some degree. This wasn’t just a paper company - they had real facilities and real operations that are now in jeopardy.

Sam Hinton: Which makes the executive departures even more concerning. If you’ve got operational facilities and paying customers, you don’t just walk away unless there are serious problems that can’t be solved.

Alex Shannon: If confirmed, this departure could signal that the nuclear-powered AI approach is harder than it looks.

Sam Hinton: Which leaves the big cloud providers in an even stronger position. They’ve got the capital and regulatory relationships to solve the energy problem at scale, while startups like Fermi apparently struggle to keep their executives in place.

Alex Shannon: And it might explain why Amazon is so confident about those circular investment deals we talked about earlier. If alternative AI infrastructure approaches keep failing, cloud providers become the only viable option for scaling AI.

Sam Hinton: Keep an eye on whether other companies pursuing similar nuclear-AI strategies start facing similar challenges.

Alex Shannon: Absolutely. And for anyone thinking about investing in alternative AI infrastructure companies, this is probably a reminder that the technical challenges are only part of the equation. Regulatory, financial, and operational challenges can be just as difficult.

Anthropic says OpenClaw-style Claude CLI usage is allowed again

Alex Shannon: Alright, let’s hit some rapid-fire stories. Anthropic has announced that OpenClaw-style Claude CLI usage is permitted once again after previously restricting it.

Sam Hinton: This is actually bigger than it sounds. CLI tools are how developers actually integrate AI into their workflows. When you restrict that, you’re basically telling developers to go use someone else’s AI.

Alex Shannon: So they realized they were shooting themselves in the foot by blocking the tools developers actually want to use?

Sam Hinton: Exactly. Plus, with that security vulnerability we just talked about, they probably want to keep developers happy while they fix the bigger issues.

Alex Shannon: It’s interesting timing though - lift restrictions on developer tools right after a major security vulnerability gets disclosed. That seems like mixed messaging about their security priorities.

Sam Hinton: Could be damage control. If developers start leaving because of the MCP vulnerability, the last thing you want is to also be blocking their preferred integration methods.

Alex Shannon: For developers who were affected by the previous restrictions, this is probably welcome news. But I’d be curious about what safeguards they’ve put in place.

Sam Hinton: Right, because they restricted it for a reason originally. Either they’ve solved whatever problem led to the restriction, or they’ve decided the problem is less important than keeping developers happy.

Atlassian enables default data collection to train AI

Alex Shannon: Early reports suggest Atlassian has enabled default data collection from its users to train AI models, making this the default setting for users.

Sam Hinton: Oh come on, Atlassian. Making data collection the default is such a sketchy move. That’s people’s work conversations, project plans, internal company documents - and now it’s AI training data unless you opt out?

Alex Shannon: This feels like the kind of decision that’s going to generate a lot of angry IT department emails.

Sam Hinton: For sure. Enterprise customers are not going to be happy about their sensitive business data being used to train models without explicit consent.

Alex Shannon: And the legal implications could be huge. If companies are using Atlassian for confidential client information or proprietary business data, they probably can’t legally consent to that being used for AI training.

Sam Hinton: Exactly. GDPR, CCPA, industry compliance requirements - there are so many regulations that prohibit using business data for purposes other than what was originally agreed to.

Alex Shannon: The fact that it’s opt-out rather than opt-in is particularly problematic. How many companies are going to realize this is happening and take action to stop it?

Sam Hinton: Probably very few, which is exactly why they made it the default. It’s the same playbook social media companies used - make data collection the default and hope people don’t notice or don’t bother to change it.

Deezer says 44% of songs uploaded to its platform daily are AI-generated

Alex Shannon: Here’s a wild stat: Early reports suggest that 44% of songs uploaded to Deezer daily are now AI-generated.

Sam Hinton: That’s actually insane. We’re not talking about a small percentage anymore - we’re talking about AI-generated music being nearly half of all new content on a major streaming platform.

Alex Shannon: What does this mean for human musicians trying to get discovered?

Sam Hinton: It means they’re competing with algorithms that can pump out songs 24/7 without needing food, sleep, or royalty payments. The economics of music creation just fundamentally changed.

Alex Shannon: And think about the quality implications. If AI can generate songs that are good enough to upload to streaming platforms, what happens to the value of human creativity in music?

Sam Hinton: That’s the scary part. Maybe AI music isn’t as emotionally resonant as human-created music, but if it’s good enough for background listening or playlist filler, that’s a huge chunk of the music market.

Alex Shannon: I’m also wondering about copyright and licensing issues. If nearly half of new uploads are AI-generated, who owns the rights to that music? The AI company? The person who prompted the AI?

Sam Hinton: That’s a legal minefield that nobody’s really figured out yet. And streaming platforms like Deezer are probably going to have to start labeling AI-generated content or face backlash from human artists.

Florida’s ‘AI Bill of Rights’: What happened, what’s in it, and what’s next? - Florida Phoenix

Alex Shannon: And finally, early reports suggest Florida has passed an ‘AI Bill of Rights’ legislation addressing artificial intelligence governance and rights.

Sam Hinton: Florida doing AI policy is not what I had on my 2026 bingo card, but honestly, good for them. Someone needs to start figuring out the legal framework for this stuff.

Alex Shannon: It’ll be interesting to see what’s actually in the bill and whether other states follow suit.

Sam Hinton: Yeah, and whether it actually has teeth or if it’s just feel-good legislation that doesn’t change how AI companies operate.

Alex Shannon: Florida’s been pretty aggressive about tech regulation lately, so I wouldn’t be surprised if this actually has some enforcement mechanisms. The question is whether it conflicts with federal AI policy or creates compliance nightmares for companies.

Sam Hinton: Right, and AI companies that operate nationally are going to hate having to comply with different state regulations. It’s like the internet privacy law patchwork all over again.

Alex Shannon: But if Florida’s bill addresses AI rights and governance, that could set a precedent for other states. We might be looking at the beginning of a state-by-state approach to AI regulation.

Sam Hinton: Which would be both good and bad. Good because someone’s finally taking AI governance seriously, bad because fragmented regulation makes it harder for everyone to comply and for consumers to understand their rights.

BIGGER PICTURE

Alex Shannon: Alright Sam, if you zoom out and look at everything we covered today - Amazon’s circular investment deals, government agencies using restricted AI despite internal conflicts, major security vulnerabilities, and executive departures - what pattern do you see?

Sam Hinton: Honestly? Chaos and consolidation happening at the same time. The big players like Amazon are using financial engineering to lock in their dominance, while the fundamental infrastructure is still fragile enough that design vulnerabilities can threaten entire supply chains.

Alex Shannon: And the government angle is particularly concerning. We’ve got different agencies apparently working at cross purposes when it comes to AI adoption and oversight.

Sam Hinton: Right, and meanwhile you’ve got half of new music being AI-generated and companies like Atlassian just deciding to hoover up user data by default. We’re in this weird phase where AI is simultaneously too powerful to ignore and too chaotic to properly regulate.

Alex Shannon: What’s really striking to me is how all these stories connect. Amazon’s betting big on Anthropic, the NSA is using Anthropic’s tools despite Pentagon concerns, but then we find out Anthropic has fundamental security vulnerabilities. It’s like everyone’s building on a foundation that might be cracked.

Sam Hinton: And that’s exactly the problem with how fast the AI industry is moving. Everyone’s making massive bets - financial, strategic, operational - without really understanding the risks. The Fermi story is a perfect example: nuclear-powered AI sounds great in theory, but the execution challenges killed it.

Alex Shannon: The data collection stories are particularly concerning when you put them together. Atlassian is defaulting to collecting business data, nearly half of music uploads are AI-generated, and government agencies are using restricted AI models. We’re seeing AI being trained on everything and deployed everywhere, often without proper oversight.

Sam Hinton: And the circular investment deals show how market concentration is accelerating. Amazon isn’t just competing in AI - they’re using their cloud dominance to create financial dependencies that lock competitors out. That’s anticompetitive behavior disguised as innovation.

Alex Shannon: Florida’s AI Bill of Rights feels almost quaint in comparison. Here’s one state trying to create some basic governance framework while federal agencies are feuding over AI use and tech giants are restructuring entire markets through circular investments.

Sam Hinton: But maybe that’s exactly what we need - someone, anyone, to start taking AI governance seriously. Even if it’s just Florida, at least it’s a start. Because the alternative is letting Amazon, Anthropic, and the NSA figure out AI policy through backdoor deals and circular investments.

Alex Shannon: What should people be watching for in the coming weeks as these stories develop?

Sam Hinton: I’d keep an eye on whether other major cloud providers start copying Amazon’s investment strategy, how quickly Anthropic can fix that security vulnerability, and whether the NSA-Pentagon tension tells us something bigger about government AI policy.

Alex Shannon: And honestly, whether Florida’s AI Bill of Rights actually does anything meaningful or just becomes a template for other states to copy.

Sam Hinton: The bigger question is whether we’re watching the AI industry mature into a regulated, stable market, or whether we’re just seeing the early stages of a much bigger consolidation where a few companies control everything and everyone else gets squeezed out.

Alex Shannon: Based on today’s stories, I’m leaning toward the latter. When investment deals are structured to create dependencies, government agencies are making unilateral AI decisions, and security vulnerabilities threaten entire supply chains, that doesn’t sound like a mature, stable market to me.

Sam Hinton: Agreed. And for anyone working in AI or depending on AI services, the lesson is probably to diversify your dependencies and have backup plans. Because when the dust settles from all this chaos and consolidation, the competitive landscape is going to look very different than it does today.

OUTRO

Sam Hinton: That’s all for today’s Build By AI. I’m still trying to process the idea that Amazon essentially paid Anthropic to become their biggest customer.

Alex Shannon: And I’m still wondering what exactly the NSA is doing with that restricted AI model. I’m Alex Shannon…

Sam Hinton: And I’m Sam Hinton. If you enjoyed today’s deep dive into AI’s weirdest week yet, make sure you’re subscribed so you don’t miss tomorrow’s episode.

Alex Shannon: See you tomorrow, and hopefully by then we’ll have some answers to today’s questions.