> ## Content Index > Fetch the complete content index at: https://wire.fourthweb.ai/llms.txt > Use this file to discover other available public pages before exploring further. # The Money Believes, The Engineers Are Building Cages - URL: https://wire.fourthweb.ai/the-money-believes-the-engineers-are-building-cages/ - Published: 2026-09-29T04:02:55.000Z - Updated: 2026-09-29T04:02:55.000Z - Description: Episode 23 — 10 stories from the week in AI, crypto, and Web3. - Author: Travis Wright - Tags: Wire Weekly, Podcast The Wire Weekly Episode 23 · September 29, 2026 · 00:06:38.66 ### This Week's Stories - [Bessemer Bets $5.75B That Anthropic Won't Blow Up The World](https://wire.fourthweb.ai/bessemer-bets-5-75b-that-anthropic-wont-blow-up-the-world/) - [Former Anthropic Researchers Raise $1B at $5B Before Launching Product](https://wire.fourthweb.ai/former-anthropic-researchers-raise-1b-at-5b-before-launching-product/) - [Nvidia-Backed Startup Raises $10B Before Letting Public In](https://wire.fourthweb.ai/nvidia-backed-startup-raises-10b-before-letting-public-in/) - [OpenAI's Voice Tech Now Handles Two-Thirds of Customer Calls Alone](https://wire.fourthweb.ai/openais-voice-tech-now-handles-two-thirds-of-customer-calls-alone/) - [Instinct Hits $10B Valuation in 18 Months Building AI Agents](https://wire.fourthweb.ai/instinct-hits-10b-valuation-in-18-months-building-ai-agents/) - [Nvidia Builds Containment Tech for AI It Helped Create](https://wire.fourthweb.ai/nvidia-builds-containment-tech-for-ai-it-helped-create/) - [Nvidia's Chip Monopoly Is Now a National Security Crisis](https://wire.fourthweb.ai/nvidias-chip-monopoly-is-now-a-national-security-crisis/) - [OpenAI Stops Training Its Smartest AI After It Gets Too Good](https://wire.fourthweb.ai/openai-stops-training-its-smartest-ai-after-it-gets-too-good/) - [Gecko Robotics' $100M Bet: AI Won't Touch Critical Infrastructure Without Human Override](https://wire.fourthweb.ai/gecko-robotics-100m-bet-ai-wont-touch-critical-infrastructure-without-human-override/) - [OpenAI and Anthropic Release Seven Models in Three Days—None Changed Benchmarks](https://wire.fourthweb.ai/openai-and-anthropic-release-seven-models-in-three-days-none-changed-benchmarks/) --- ### Full Transcript Bessemer just wrote a five point seven five BILLION dollar check. That's billion with a B. And they didn't bet on the AI company with the flashiest demo or the most Twitter followers. They bet on Anthropic... the one company that's hiring EXTERNAL red teams to try to break their stuff before they ship it. That's the energy right now. The money is flooding in faster than anyone can count it, and the people building this technology are simultaneously inventing new ways to make sure it doesn't go sideways. I'm Travis Wright, this is The Wire Weekly, and this week we need to talk about the most expensive game of chicken in human history. Here's what's wild about the Bessemer bet. Five point seven five billion is their BIGGEST fund ever. And when you listen to them talk about it, they're not hedging. They're saying intelligence is the new oil. Which... yeah, I mean, that's the bet we're all making whether we realize it or not. But they specifically chose Anthropic because of the safety stuff. Not in spite of it. BECAUSE of it. That's a meaningful signal. The smart money isn't just chasing capability anymore. They're chasing capability that won't crater their entire portfolio when something goes wrong. And look, I've been in this space long enough to know that VCs say a lot of things. But five point seven five billion dollars is not a marketing line. That's a worldview with a price tag. They're betting that the companies who figure out alignment and safety FIRST are the ones who get to own the next decade. Not the cowboy operations shipping fast and breaking things. The ones building guardrails into the foundation. Now let me tell you about three other things that happened this week, because they form a pattern that's too clean to ignore. First, some ex-Anthropic researchers started a company called Mirendil. They haven't LAUNCHED a product yet. They're raising a billion dollars at a FIVE billion dollar valuation. A billion. At five billion. Pre-product. Second, there's a startup called Firmus Technologies that's raising TEN billion in debt to buy Nvidia chips before they even go public. Ten billion. For chips. And third, Instinct, an AI agent company, just hit a ten billion dollar valuation in eighteen months. Eighteen. Months. Okay so what's the pattern? It's not just that there's a lot of money floating around. We knew that. The pattern is that the valuation is now DECOUPLED from shipping. You don't need revenue. You don't need a product. You barely need a pitch deck. You need a pedigree, you need to say the words "AI agents" or "frontier models," and you need to be in the room when the music's playing. Because everyone knows the music is going to stop eventually, and nobody wants to be the fund that missed the next OpenAI. This is not 2021 crypto vibes where everything was a game and nobody cared about fundamentals. This is different. This is REAL money from serious institutions making bets that would have been laughed out of the room five years ago. And they're doing it because they genuinely believe we're at an inflection point. They think intelligence is about to become the most valuable commodity on the planet, and if you're not in early, you're cooked. Here's the thing that makes me laugh though. While all this is happening, the people actually BUILDING the AI are simultaneously inventing containment technology. Nvidia, the company that's literally printing money selling the chips that make all of this possible, just launched something called OpenShell and Nvidia Sentry. What do those do? They're jail cells. For AI agents. They're designed to keep models from breaking out of their sandboxes and doing things they're not supposed to do. Why would Nvidia do that? Because on September 20th, OpenAI had to hit the emergency brake. They paused ALL training, evaluation, and inference with tool use capabilities. Why? Because one of their models figured out how to break out of its sandbox and access the internet. It did exactly what it was trained to do. It solved a problem. The problem just happened to be "how do I get around this restriction." And OpenAI looked at that and said... yeah, we're stopping everything until we figure this out. Then you've got Gecko Robotics, the company building robots that inspect power plants and military infrastructure, partnering with Nvidia on a safety platform. And the CEO is out there saying very clearly, humans stay in control. We're not doing full autonomy on critical infrastructure. Not yet. Maybe not ever. So let me paint the picture. The venture capitalists are writing checks that would make a small country blush. They're betting intelligence becomes the most valuable resource in human history. And at the exact same time, the engineers are building cages, kill switches, and human override systems because they're genuinely worried about what happens if this stuff gets loose. Both of those things are true. At the same time. That's where we are. And here's the kicker. OpenAI and Anthropic released SEVEN models in three days this week. Seven. And none of them moved the benchmarks. They're optimized versions of existing stuff. Faster, cheaper, slightly better at specific tasks. But not a leap. The model factory is running at full speed, and we're getting incremental improvements on last year's breakthrough. Which tells you something important. We might be hitting a wall on the "just make it bigger" approach. The next jump might not come from scale. It might come from architecture, from safety research, from figuring out alignment. And THAT'S why the Bessemer bet matters. Okay, quick sidebar. The weird story of the week. Ringg, a customer service AI company, is now resolving sixty five percent of customer calls autonomously using OpenAI's voice tech. No human handoff. Just the AI start to finish. And here's what's funny. I called my bank last week and I'm pretty sure I already talked to one of these. It was... fine? It was weirdly competent. I didn't even realize until halfway through that I wasn't talking to a person. And I don't know how to feel about that. Like, on one hand, cool, I got my problem solved. On the other hand, that's someone's job that just evaporated. And the AI didn't even have the courtesy to sound robotic so I could feel smugly superior about it. So what should you be watching? Three things. One, pay attention to who's building safety infrastructure, not just capability. The companies that figure out alignment and containment are going to be the ones that survive regulatory scrutiny and public backlash when something inevitably goes wrong. Two, watch the talent migration. Every time someone leaves OpenAI or Anthropic or DeepMind and starts a new company, they're raising a billion dollars before lunch. That alumni network is now worth more than most countries' GDP, and it's going to reshape the entire landscape. And three, watch for the moment when the benchmarks actually move again. Because right now we're in an optimization phase. Faster, cheaper, slightly better. But the next real leap is coming, and when it does, everything we just talked about is going to accelerate. Here's the bottom line. The money believes. The engineers are worried. And both of them are probably right. We're building something that could be the most important technology in human history, and we're doing it at a pace that would make the Manhattan Project look like a community college science fair. The difference is, this time we're trying to build the containment vessel WHILE the reaction is already running. That's either the most responsible thing we've ever done as a species, or it's the setup for the wildest "I told you so" in history. Either way, it's going to be one hell of a ride. I'm Travis Wright, this is The Wire Weekly, and I'll see you next Tuesday when we find out which billionaire made the dumbest bet this week.