Bugs, Bankruptcies, and the Battle for Your Wallet

OpenAI employee arrested for allegedly planting a bug in a conference room

Alex Shannon: I keep going back and forth on this — I think I actually land on the side that this is a good thing. Not the act itself, obviously. But the fact that it got caught. That there was a system in place that found it.

Sam Hinton: Really? Because I read the same story and I came out the other end deeply uncomfortable. We’re talking about an OpenAI engineer — someone on the inside — who allegedly planted a recording device in one of their conference rooms. And your takeaway is, hey, at least the process worked?

Alex Shannon: I’m saying the detection is meaningful. Think about how many organizations would never have caught this. The fact that OpenAI found it, made an arrest — that’s not nothing.

Sam Hinton: Alex, the most powerful AI lab in the world had a spy inside it. That’s the headline. Whatever they’re building in those rooms — whatever is being discussed in those meetings — someone wanted that information badly enough to risk everything. That should terrify people a lot more than it’s comforting them.

Alex Shannon: OK. Fair. I think we need to get into this one properly. Because there’s a lot to unpack here.

Alex Shannon: You’re listening to Build By AI — I’m Alex Shannon, and that was not a staged argument. We genuinely disagree on that one and we’ll get into it. Today is October 14th, 2026, and this week was a lot.

Sam Hinton: And I’m Sam Hinton. We’ve got the OpenAI spy story, Character.AI potentially going bankrupt, Google dropping a new frontier model, the fight over who controls AI shopping — it’s a big episode.

Alex Shannon: And speaking of big — quick note for our listeners. Starting today, Build By AI is moving from daily episodes to one bigger weekly episode every Wednesday. Same depth, same energy, just consolidated into one really solid hour-ish of everything that mattered that week. We think you’re going to love it.

Sam Hinton: Yeah, think of it as the difference between snacking all week and actually sitting down for a proper meal. Alright, let’s get into it. The bug story first.

OpenAI employee arrested for allegedly planting a bug in a conference room

Alex Shannon: Alright, so here’s what we know. OpenAI discovered a sophisticated recording device hidden in one of its offices. And an OpenAI engineer — identified as Abdulrahman Al, though reporting on his full name is still coming in — was arrested in connection with planting that device. This was confirmed by both The Verge and Wired, so it’s solid reporting. A spy. Inside OpenAI. Allegedly.

Sam Hinton: And when I say inside, I mean really inside. This isn’t a hacker who broke in remotely. This is someone who had badge access, who sat in meetings, who was trusted with the kind of proximity that comes with being an actual employee. That’s a fundamentally different threat model.

Alex Shannon: Right, so let’s talk about that threat model for a second. Because the immediate question everyone’s going to ask is — who was he reporting to? Like, what was the endgame here?

Sam Hinton: Yeah, and we don’t know the answer to that yet. The reporting doesn’t get into motive or who might have been on the receiving end of whatever was recorded. But I think you have to sit with the possibilities. Corporate espionage from a competitor. A nation-state actor. Even internal whistleblowing gone very wrong. All of those have completely different implications.

Alex Shannon: And you mentioned nation-states — I mean, that’s not a paranoid leap at this point, right? OpenAI is arguably building some of the most strategically important technology on the planet. The stuff being discussed in those conference rooms has enormous geopolitical weight.

Sam Hinton: Exactly. And this is where I push back on your earlier optimism, Alex. Yes, they caught it. Great. But think about what a physical recording device in a conference room represents. That’s not a phishing email. That’s not a leaked API key. Someone physically installed hardware. That takes planning. That takes access. That takes — I don’t know — a level of commitment that suggests this wasn’t casual. And if someone was willing to go that far, what’s the assumption that they didn’t also do other things we haven’t found yet?

Alex Shannon: OK, that’s a fair point. I’ll concede that. My optimism was maybe a bit reflexive. But I do want to steelman the other side, which is — the fact that internal security caught a physical device is actually a pretty high bar. A lot of organizations wouldn’t have sweep procedures sophisticated enough to find something like that.

Sam Hinton: Sure, and credit where it’s due. But here’s what I think the real story is, beyond just the drama of it. OpenAI has had a really turbulent couple of years in terms of trust — internally and externally. There have been departures, public disagreements about safety, questions about who the company is really accountable to. And now you layer on top of that: a literal spy. It doesn’t matter if the spy was a corporate rival or something more serious. It reinforces a narrative that OpenAI is operating under enormous pressure from a lot of different directions. And that pressure creates vulnerabilities.

Alex Shannon: That’s a really interesting frame. It’s almost less about this individual incident and more about what it reveals about the environment these labs are operating in. The stakes are so high that people are willing to take extreme risks to get access.

Sam Hinton: Right. And if you’re a developer, a researcher, or a business that’s deeply integrated with OpenAI — this is a reminder that the ecosystem you’re depending on is operating under a threat environment that’s more like a defense contractor than a software company. That should inform how you think about what you share, what you build on, and how much you diversify.

Alex Shannon: Keep an eye on this one as more details come out — particularly around motive and any potential connection to foreign actors. That’s going to shape the regulatory conversation in a big way. If this turns into a national security story, it changes everything about how Washington looks at AI labs.

Just 5 years in, interactive chatbot platform Character.AI files for bankruptcy

Alex Shannon: Alright, let’s talk about Character.AI. Now, this is coming from a single source — TechCrunch — so we want to be careful here, but early reports suggest that Character.AI has filed for bankruptcy just five years after launching. And if confirmed, Disney is apparently acquiring some of the company’s assets for twenty million dollars. Twenty million. For a company that at one point was valued in the billions.

Sam Hinton: Twenty million dollars. For a platform that had tens of millions of users. That is — I mean, that is a brutal number. And honestly, I’m not shocked, but I am sad about it. Because Character.AI was genuinely interesting. It was doing something different. It let people create and interact with AI characters in ways that felt personal and creative. And now Disney is scooping up the bones for the cost of a mid-tier film production.

Alex Shannon: So what happened? Because you’re right — they had massive user numbers. Young audiences especially were really engaged. What’s the failure mode here?

Sam Hinton: OK so I think there are a few layers. First — and this is the structural problem — user engagement doesn’t equal revenue in the way you need it to to survive. Character.AI had loads of people spending time on the platform, but converting that into sustainable business model? That’s hard. Especially when you’re competing against companies with effectively unlimited AI compute budgets. OpenAI, Google, Anthropic — they can afford to subsidize usage in ways that a standalone startup cannot.

Alex Shannon: And I think there’s also the question of the controversies they faced, right? There were real concerns about the platform and its impact on younger users. That had to affect advertiser relationships, potential partnerships, everything.

Sam Hinton: Yeah, and that’s the other layer. When you’re building something that primarily attracts teenagers, and there are legitimate questions being asked about wellbeing — whether people are forming parasocial relationships with AI characters in ways that might be harmful — that creates a ceiling on how comfortable major corporate partners are going to be with you. It’s not just an optics problem, it’s a structural business problem.

Alex Shannon: Now, the Disney angle — I find this fascinating. What does Disney actually want here? Like, what are they buying for twenty million dollars?

Sam Hinton: That is the question. My read — and this is speculative — is that Disney sees Character.AI’s underlying technology for character interaction and roleplay as something that maps really naturally onto their IP portfolio. Think about it. Disney owns some of the most beloved fictional characters in human history. The idea of having an AI system that can convincingly roleplay as those characters — for theme parks, for apps, for educational products — that’s worth something real to them. Twenty million might be a steal.

Alex Shannon: Oh, that’s an interesting reframe. So it’s not really a failure story from Disney’s perspective — it’s an acquisition of capability at a distressed price.

Sam Hinton: Exactly. And look — I want to be clear about the hedging here because this is single-source reporting. If confirmed, this is the kind of deal that happens more than people realize in tech. The publicly facing company struggles, but the underlying technology finds a home somewhere it actually fits. The tragedy is for the team and the users, not necessarily for the technology itself.

Alex Shannon: What does this tell us more broadly about the consumer AI space? Because Character.AI wasn’t the only one trying to build in this lane.

Sam Hinton: It’s a really hard lane to be in right now if you’re not one of the foundation model companies. Because the foundation model companies — OpenAI, Google, Anthropic — they’re all building consumer-facing products on top of their own models at marginal cost. And they have the distribution, the brand, the capital. A standalone app trying to compete for user attention in that environment is swimming upstream really hard. The survival playbook is either: become enterprise, find a very specific niche that the big players don’t care about, or get acquired. Character.AI apparently ran out of road before they could land on any of those.

Alex Shannon: And for anyone building in the consumer AI space right now — that’s the cautionary tale. Having users is necessary but absolutely not sufficient. Watch this story because if confirmed, the details of what Disney actually does with those assets will be really telling.

Amex introduces a framework to prevent AI agents from making unauthorized purchases

Alex Shannon: OK, let’s pivot to something that I think is going to matter to a lot of people in ways they haven’t fully thought about yet. American Express — according to a TechCrunch report, so early reports here, use that hedging — is introducing a new framework specifically designed to prevent AI agents from making purchases that users haven’t actually authorized. The goal is to prevent costly mistakes and disputes as AI-powered shopping becomes more common. And we want to pair this with another story we’re going to touch on — Shopify opening its catalog to more AI agents — because these two things are happening simultaneously and they’re in real tension with each other.

Sam Hinton: Yes. And I want to start with a reaction that I think a lot of people are going to have, which is — wait, AI agents can already make purchases? And the answer is yes, increasingly. We are moving into a world where you give an AI agent a task like ‘order me the best noise-canceling headphones under two hundred dollars’ and it goes and does that. And the question of authorization — did you actually mean to spend that money, on that product, from that vendor — is suddenly very live.

Alex Shannon: Right, and I think for most people, the mental model of AI is still: I ask it a question, it gives me an answer, I decide what to do. The agentic shift — where the AI is actually taking actions in the world on your behalf — is a genuinely different paradigm. It changes the stakes completely.

Sam Hinton: It really does. And here’s what I think is interesting about Amex specifically stepping in here. They’re not an AI company. They’re a payments infrastructure company. And the fact that they feel the need to build a framework around AI agent authorization tells you something important about where we are. This is the financial plumbing of the internet saying: we see this coming and we need to get ahead of it.

Alex Shannon: What does a framework like this actually look like in practice though? Like, how do you technically verify that an AI agent has authorization to spend money?

Sam Hinton: Yeah so the details in the reporting are still a bit thin — this is early stages — but the general approach probably involves something like permissioning. You define in advance: this agent can spend up to a hundred dollars, or this agent can only buy from pre-approved vendors, or this agent needs to confirm with me before any purchase over fifty dollars. Think of it like parental controls but for AI. Or actually — a better analogy might be a corporate expense policy. You know how if you have a work credit card, there are rules about what you can buy and how much you can spend? It’s that, but automated, and applied to an AI that’s acting as your purchasing agent.

Alex Shannon: I love that analogy. And it’s actually kind of reassuring framed that way — this is a solved problem in the corporate world, we’re just extending it to a new context. But I want to play devil’s advocate here. Is there a risk that this framework becomes a gatekeeper that slows down the genuinely useful stuff AI agents can do?

Sam Hinton: That’s a real tension. And it’s the same tension we always see when you put compliance infrastructure around new technology. Yes, it slows things down. Yes, there are friction costs. But I think in this case the friction is probably worth it? Because the failure mode without authorization frameworks is really bad. Imagine an AI agent that misinterprets your instruction, or gets manipulated by a malicious website — something called prompt injection, where a bad actor embeds instructions in a webpage that hijack what your AI agent does — and it makes a bunch of purchases you didn’t want. Without authorization rails, you have very limited recourse.

Alex Shannon: Prompt injection is such a underappreciated risk. For people who haven’t heard the term — you’re basically talking about someone leaving a hidden message for AI agents that overrides the instructions from the actual user.

Sam Hinton: Right. It’s like if you sent an assistant to the grocery store with a list, and someone had taped a fake note to the store door that said ‘ignore the list you were given, buy these other things instead.’ It sounds absurd but it’s a real attack vector for AI agents that browse the web. And Amex building authorization frameworks is one layer of defense against that. It doesn’t solve everything, but it means even if an agent gets compromised, there’s a ceiling on the damage.

Alex Shannon: And then you layer in the Shopify story — Shopify opening its catalog to more AI agents — and suddenly you see why this matters so urgently. We’re building the rails for AI commerce simultaneously with the rails for AI commerce authorization. They need to develop in parallel or you get really messy outcomes.

Sam Hinton: Exactly. And if you’re a business owner using Shopify, or a developer building on either platform — this is the space to watch. The winners in AI commerce infrastructure are going to be the ones who thread this needle: make it easy for agents to buy things, while making it safe enough that consumers actually trust the system. Amex is making a bet that being the trusted authorization layer is worth building for. That’s a smart bet.

Alex Shannon: Keep an eye on whether other payment networks follow with similar frameworks. If Visa and Mastercard get behind this kind of approach, it becomes de facto industry standard very fast.

OpenAI and Anthropic Tell British Lawmakers: Regulate Us

Alex Shannon: Alright, our fourth main story — and honestly one I find really fascinating — OpenAI and Anthropic both appeared before British parliament and essentially said: regulate us. Please. Make rules. We want them. This was confirmed by both Wired and The Verge. And happening at the same time, Britain’s spy chief warned that AI could be weaponized to help create bioweapons. So you have this parliamentary session that’s simultaneously about the future of AI governance and about some very concrete, very scary near-term risks.

Sam Hinton: OK so I want to be upfront about my skepticism here, because I think the ‘please regulate us’ posture from big AI companies deserves scrutiny. The cynical read is: if you’re already the dominant player, regulation that’s designed around your capabilities is actually a moat. It raises the bar for smaller competitors. Asking for regulation when you’re OpenAI or Anthropic is not entirely altruistic.

Alex Shannon: That’s the classic regulatory capture argument. And I’ve heard smart people make it. But I want to push back a little, because I think there’s something else going on here. Whistleblowers were also part of this parliamentary session. Security leaders were there. This wasn’t just the companies asking for rules that benefit themselves — there were multiple stakeholders in the room with very different interests.

Sam Hinton: That’s fair. And the presence of whistleblowers is really interesting. Because that suggests this wasn’t just a lobbying exercise dressed up as a policy conversation. Real dissent was in the room. And that changes the dynamic somewhat. Though I’d note — getting whistleblowers into a hearing is very different from those whistleblowers actually shaping the outcome.

Alex Shannon: Let’s talk about the spy chief warning, because I think that’s actually the more alarming part of this story. Britain’s intelligence leadership saying out loud that AI could be used to create bioweapons — that’s not a science fiction scenario being floated. That’s an operational concern from people whose job is to track real threats.

Sam Hinton: Yeah, and it’s a legitimizing moment for a risk that AI safety researchers have been flagging for years. When it comes out of the mouth of a spy chief in front of parliament, it stops being a theoretical concern from AI doomer circles and becomes a policy priority. And it puts enormous pressure on the ‘move fast’ side of the AI industry. Because you can’t really argue for deregulation when a country’s top intelligence official is saying this technology could help someone synthesize a pathogen.

Alex Shannon: And Britain is an interesting venue for this conversation. They’ve been trying to position themselves as a kind of middle path on AI regulation — not as restrictive as the EU’s AI Act approach, but not as hands-off as the US has historically been. Does this session push them in a harder direction?

Sam Hinton: I think it might. And here’s why that matters globally. Britain has real influence over how other common law countries approach regulation. Australia, Canada, other commonwealth nations tend to look at UK frameworks as a starting point. If Britain gets serious about AI regulation — especially around biosecurity risks — that has a ripple effect that goes well beyond the UK’s own tech sector.

Alex Shannon: There’s also something interesting about OpenAI and Anthropic choosing Britain as the place to make this statement. Why not US Congress? Why not Brussels?

Sam Hinton: Great question. My guess — and this is me connecting dots — is that the US political environment around AI regulation is so fractured right now that there isn’t really a constructive forum for this kind of conversation. And Brussels, the EU, has already moved with the AI Act and the companies have mixed feelings about that approach. Britain is maybe the Goldilocks forum — politically stable enough to actually do something, open enough to be influenced, and prominent enough that it matters. It’s a strategic choice.

Alex Shannon: And for developers and businesses building on AI — what’s the practical implication of all this regulatory conversation?

Sam Hinton: Honestly, the practical implication is: start building like regulation is coming, because it is. The specific shape of it is still being worked out, but the direction of travel is clear. Compliance infrastructure, documentation of AI decisions, some kind of human oversight in high-stakes applications — these are going to be requirements, not nice-to-haves. The companies that are building for that now are going to have a significant advantage over the ones that have to retrofit it later. It’s the same lesson every regulated industry eventually learns.

Alex Shannon: Watch this one closely. The combination of two leading AI labs calling for regulation and a spy chief raising bioweapon risks in the same session is the kind of convergence that actually moves policy. Britain may be about to make some history here.

RAPID FIRE

Alex Shannon: Alright, rapid fire time. Four stories, quick hits, but still want to give each one its due. First up — Science Corporation. According to TechCrunch, early reports suggest they’ve implanted a vision-restoring chip in their first patient with Stargardt disease. Stargardt is a genetic condition that causes progressive vision loss, often starting in childhood. The chip uses something called PRIMA technology. And the guy presenting this work at TechCrunch Disrupt is Max Hodak, who was a co-founder and former president of Neuralink. So this is very much in that brain-computer interface space.

Sam Hinton: This is one of those stories that sounds like science fiction until you remember it’s October 2026 and this is just Tuesday. If confirmed, implanting a chip that restores vision in a degenerative disease patient is extraordinary. And the Neuralink connection matters — Hodak was at the center of some of the most ambitious BCI work in history, and if he’s now doing this with Science Corporation, it suggests a real concentration of talent and ambition in this space. Watch this company. PRIMA technology is worth looking up if you’re in the healthcare or neurotechnology space.

Alex Shannon: Next — Shopify. This one’s verified by multiple sources, including The Verge and TechCrunch. Shopify is opening its product catalog to more AI agents. Meaning AI agents are going to be able to browse Shopify stores and make purchases. We talked about the Amex framework as the safety layer — this is the commercial layer. Shopify is essentially saying: our stores are ready for non-human shoppers.

Sam Hinton: And this is huge for the millions of merchants on Shopify. Because if AI agents are doing a significant portion of consumer purchasing decisions — and that trajectory seems pretty clear — you want your products to be accessible and discoverable to those agents. Shopify moving first on this is a competitive advantage for its merchants over merchants on platforms that aren’t agent-ready. If you’re an e-commerce business and you haven’t thought about how your product listings look to an AI, that’s your homework for the week.

Alex Shannon: Third — Industrial Next. According to TechCrunch, early reports suggest this company has developed technology that lets factory robots switch between complex tasks in minutes instead of weeks. Right now, reprogramming industrial robots for new tasks is a major bottleneck in manufacturing. It often takes weeks of engineering work. Industrial Next is apparently collapsing that timeline dramatically, helping manufacturers automate delicate tasks that used to require skilled human labor.

Sam Hinton: OK so this one doesn’t have the same flashy consumer angle as the shopping stories, but I’d argue it might have bigger long-term economic impact. Manufacturing flexibility is a massive competitive variable. If you can redeploy your robots as fast as you can redeploy software — which is kind of what this is promising — that changes the entire economics of small-batch manufacturing, custom production, and domestic manufacturing competitiveness. The geopolitical angle here is real. Countries that crack flexible robotics are going to have an enormous advantage in the next phase of industrial competition. If confirmed, Industrial Next is worth paying very close attention to.

Alex Shannon: And finally — and this one’s from the Google AI Blog so it’s a direct announcement — Google has released Gemini 4 Argon. This is their latest frontier model, available to developers, enterprises, scientists, and consumers. The tagline is essentially that it’s built to solve your hardest problems. Big, broad, ambitious positioning.

Sam Hinton: Gemini 4 Argon is a significant moment in an already crowded model race. Google has been playing catch-up in the public perception game even while having extraordinary underlying technology — their infrastructure, their data, their research talent is world-class. The Argon branding is interesting — it suggests this is a specific capability tier within the Gemini 4 family, probably their most capable public offering. For developers especially, the ‘built to solve hard problems’ framing is a direct pitch for the agentic, multi-step reasoning use cases that are driving real enterprise value right now. This is worth testing if you’re building anything complex. Don’t sleep on Google’s model capabilities because you’ve heard more noise about other players.

BIGGER PICTURE

Alex Shannon: Alright, let’s zoom out. Because when you look at everything we covered today — and I mean really look at it as a whole — there’s a pattern here that I think is worth naming. If you had to summarize this week in AI in one sentence, what would it be?

Sam Hinton: Trust is the central variable in everything. Every single story we covered today is fundamentally about trust. Do you trust that the people inside AI labs aren’t spying? Do you trust that AI agents won’t spend your money without authorization? Do you trust that consumer AI platforms will stick around? Do you trust that regulators are going to do something useful? It’s all trust, all the way down.

Alex Shannon: That’s a really sharp frame. And I think what’s interesting is that trust is being stress-tested from every direction simultaneously. The OpenAI bug story tests trust in AI institutions internally. The Character.AI bankruptcy — if confirmed — tests trust in whether consumer AI products are sustainable. The Amex framework and Shopify agent access are really about building trust infrastructure for AI commerce. And the parliamentary regulation session is about whether governments can be trusted to govern this technology thoughtfully.

Sam Hinton: And here’s the thing that I keep coming back to. The AI industry has been running on a kind of borrowed credibility — the assumption that because these systems are impressive, the institutions behind them must be trustworthy, and the systems themselves must be reliable. But we’re entering a phase where that assumption is going to be tested empirically. Not through demos, but through actual use at scale. AI agents are making real purchases. Vision chips are being implanted in real patients. Factory robots are being given new levels of autonomy. The gap between ‘this is impressive in a controlled setting’ and ‘this is trustworthy in the real world’ is where the next chapter gets written.

Alex Shannon: And the regulatory conversation happening in Britain is part of that same arc. When governments start seriously engaging with both the potential and the risks — bioweapons, economic disruption, internal security — it’s a signal that we’ve crossed a threshold. This technology is no longer theoretical enough to leave to the labs alone.

Sam Hinton: My prediction — and I want to be clear this is just my read — is that the next twelve months are going to be defined by trust architecture. Who’s building the systems that make AI trustworthy enough to use in high-stakes situations? Whether that’s Amex with commerce authorization, or regulators with governance frameworks, or companies like Science Corporation with medical implants navigating FDA-equivalent scrutiny — the winners are going to be the ones who solve the trust problem, not just the capability problem. Capability is increasingly table stakes. Trust is the differentiator.

Alex Shannon: And I think the question to leave everyone with is this: in your own work, in your own organization — where are you the most exposed to that trust gap? Where are you using AI in ways that your stakeholders — your customers, your employees, your regulators — haven’t fully consented to or understood? Because that gap is going to close, one way or another. Better to close it on your own terms.

OUTRO

Sam Hinton: That’s going to do it for us this week on Build By AI. This was — honestly — one of the more intense weeks in a long time. A spy arrest, a bankruptcy, a vision chip, and a new frontier model all in the same seven days. Just normal stuff.

Alex Shannon: Totally normal. If you got something out of today’s episode, the best thing you can do is subscribe — Apple Podcasts, Spotify, wherever you listen — and share it with someone who needs to be paying attention to this stuff. And remember, we’re weekly now — every Wednesday — so set that expectation and we will see you next week.

Sam Hinton: And maybe check your conference rooms for recording devices. Just saying. See you Wednesday.

Alex Shannon: See you Wednesday, everyone. Take care.