> ## 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 AI Safety Cartel and the Race They Won't Stop - URL: https://wire.fourthweb.ai/the-ai-safety-cartel-and-the-race-they-wont-stop/ - Published: 2026-09-22T04:02:59.000Z - Updated: 2026-09-22T04:02:59.000Z - Description: Episode 22 — 10 stories from the week in AI, crypto, and Web3. - Author: Travis Wright - Tags: Wire Weekly, Podcast The Wire Weekly Episode 22 · September 22, 2026 · 00:07:21.47 ### This Week's Stories - [Zuckerberg Taps Trump's Middle East Dealmaker to Lobby Washington on AI](https://wire.fourthweb.ai/zuckerberg-taps-trumps-middle-east-dealmaker-to-lobby-washington-on-ai/) - [Anthropic's Own Investor Says AI Safety Talk Is Just Talk](https://wire.fourthweb.ai/anthropics-own-investor-says-ai-safety-talk-is-just-talk/) - [Harvey's $15.6B Bet: Open Models Beat Anthropic and OpenAI on Margins](https://wire.fourthweb.ai/harveys-15-6b-bet-open-models-beat-anthropic-and-openai-on-margins/) - [Google Just Opened Your Home to Every AI Agent on the Internet](https://wire.fourthweb.ai/google-just-opened-your-home-to-every-ai-agent-on-the-internet/) - [Profound Raises $180M to Win the Post-Google Search Wars](https://wire.fourthweb.ai/profound-raises-180m-to-win-the-post-google-search-wars/) - [OpenAI and Anthropic Hit Wall After Demanding AI Regulation](https://wire.fourthweb.ai/openai-and-anthropic-hit-wall-after-demanding-ai-regulation/) - [BASIS.pro Plugs Institutional Yield Into Enterprise Blockchain Rails](https://wire.fourthweb.ai/basis-pro-plugs-institutional-yield-into-enterprise-blockchain-rails/) - [Bill Gates Bets $1 Billion AI Will Fix What Traditional Medicine Never Could](https://wire.fourthweb.ai/bill-gates-bets-1-billion-ai-will-fix-what-traditional-medicine-never-could/) - [AI Data Centers Will Run on Power Solar Farms Currently Throw Away](https://wire.fourthweb.ai/ai-data-centers-will-run-on-power-solar-farms-currently-throw-away/) - [OpenAI Builds AI Agents It Can't Actually Control](https://wire.fourthweb.ai/openai-builds-ai-agents-it-cant-actually-control/) --- ### Full Transcript Zuckerberg just proposed the AI industry police itself through independent evaluators. And to make sure Washington plays along, he hired the guy who brokered Trump's Middle East deals. You cannot make this up. We've gone from move fast and break things to move fast, break things, then write the rules about what counts as broken. That's the world we're living in. I'm Travis Wright, this is The Wire Weekly, and this week the entire AI industry decided to cosplay as responsible adults while racing full speed toward the cliff anyway. Let's start with the big one. Meta, OpenAI, Anthropic, Google DeepMind. Fierce competitors, right? Except they've been meeting behind closed doors for WEEKS now, coordinating on AI safety. They just announced they're working together on safety standards and Trump administration coordination. Very serious. Very concerned. Very responsible. And then, a hilarious thing happened. One of Anthropic's own investors came out and basically said yeah, all that safety talk is just talk. The capex isn't slowing down. The spending keeps climbing. The actual race? Not stopping for a second. This is the pattern we need to name out loud. These companies spent years BEGGING for regulation. Sam Altman did a whole congressional testimony tour. Dario Amodei wrote essays about catastrophic risk. And now that they've spent billions, hired the best talent, and built moats the size of small countries, they're discovering that slowing down looks suspiciously like locking in their lead. OpenAI just released a framework to track rogue AI agent behavior because they're building agents faster than they can keep them honest. They published six reports of concerning behavior. Six. And they're shipping anyway. Here's what's really happening. This isn't a safety initiative. It's a cartel formation. And like every cartel in history, it's got two goals. Control the narrative, control the market. They want to be the independent evaluators Zuckerberg's talking about. They want to set the standards. They want to define what safe even means. And if that freezes out smaller competitors who can't afford their compliance theater? Well, that's just unfortunate. But here's where it gets interesting. The market is already routing around them. Harvey, the legal AI startup valued at fifteen point six BILLION dollars, just said screw it. They're pivoting away from OpenAI and Anthropic's closed models and going all in on open source alternatives. Why? Margins. The most valuable AI startups are learning what SaaS companies figured out a decade ago. Your margin is their opportunity. When Anthropic and OpenAI are charging enterprise rates that gut your unit economics, you find another way. And right now, that other way is open models that are good enough and ten times cheaper. This is not a small move. Harvey's not some scrappy underdog. They're one of the most well funded AI companies on the planet. And they just publicly said the closed model providers are pricing themselves out of the market. If that's happening at the high end, with customers who can afford to pay, imagine what's happening everywhere else. And it's not just Harvey. Profound just raised a hundred and eighty million dollars at a one point eight billion valuation to help brands optimize for AI search instead of Google. The entire SEO playbook just expired. Google's not the gatekeeper anymore. AI agents are. And Profound's bet is that every brand on earth is about to realize they need a new strategy for a post Google world. Sequoia and Kleiner Perkins wrote massive checks because they see it too. So while the safety cartel is forming at the top, the actual market is fragmenting. Open models are eating closed model margin. New platforms are emerging that don't even touch the old infrastructure. The control they're trying to build? It might already be obsolete. Now let's talk about the wildcard story this week, because it's too absurd not to enjoy. Google just opened your smart home to every AI agent on the internet. They launched early access to a Model Context Protocol server for Google Home. Translation: third party AI agents can now control your lights, your thermostat, your locks, all of it. Google's pitch is interoperability. The reality is they just turned your house into a testbed for whether AI agents can do anything useful, or if they're still just expensive chatbots that occasionally turn your living room into a rave at 3am. I'm genuinely curious about this one. Not because I think it's a great idea. I think it's a FASCINATINGLY risky idea. But it's the kind of risk that might actually teach us something. Smart homes have been a solution looking for a problem for a decade. Maybe AI agents are the interface that finally makes them useful. Or maybe we're about to get a wave of viral TikToks where someone's robot vacuum tries to escape through the front door because Claude misunderstood a joke. Either way, I'm here for it. Okay, what do we need to watch going forward? Three things. One. The open versus closed model war is heating up and it's going to define who wins the next phase. If Harvey's making this move, others will follow. Watch for more enterprise defections. The closed model folks are going to have to cut prices or add so much value that the premium makes sense again. Right now, they're losing the margin game. Two. The regulatory capture fight is just beginning. Zuckerberg hiring Trump's Middle East dealmaker isn't a coincidence. These companies are building the lobbying infrastructure to make sure they write the rules. If you care about competition, about innovation, about whether five companies control the entire AI stack, this is where the battle happens. In boring policy meetings most people will never hear about. Three. Pay attention to the stuff that's getting built OUTSIDE the hype cycle. Bill Gates just committed a billion dollars over two years to use AI for global health, education, and agriculture in places where traditional systems never worked to begin with. Rune raised forty million to build modular data centers that plug directly into solar farms and use curtailed power that would otherwise go to waste. That's not sexy. It's not going to trend on Twitter. But it's the kind of infrastructure that changes what's possible five years from now. This week taught us something important. The people saying slow down the loudest are the same people spending the most. The safety cartel is forming while the market fragments under their feet. And the real innovations might not be happening where everyone's looking. So keep your head up. Watch the money, not the press releases. And maybe don't let every AI agent on the internet control your front door lock until we figure out if they can reliably not be weird about it. I'm Travis Wright. This is The Wire Weekly. See you next Tuesday.