Bots Own the Internet Now

Yesterday in AI - Weekend Recap | Monday, June 8, 2026

Bots Own the Internet Now

The internet passed a milestone this weekend that nobody threw a party for: automated traffic officially crossed 57.2% of all web traffic, meaning bots now outnumber humans online for the first time in history. AI agents are forming their own category of user, autonomous, adaptive, and increasingly indistinguishable from the real thing. We dig into what that means for every website, every API, and every digital product built on the assumption of a human at the other end.

We also get into OpenAI's plan to kill ChatGPT as we know it and relaunch it as a "super app" bundling Codex, agents, and a personal assistant layer, with one senior employee apparently telling colleagues that "Chat is dead." Plus: why the S&P 500's refusal to fast-track SpaceX could slow down the OpenAI and Anthropic IPOs, 150 mathematicians signing a declaration against AI math hype, a University of Toronto team building a self-spreading AI worm from free public models, and the school shooting lawsuit that just became the most important AI product liability case in the country.

Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Transcript

SPEAKER_00 0:05

Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of AI in 10 minutes or less. I'm Mike Robinson. It's Monday, June 8th, and something quietly crossed a threshold in this weekend's coverage that we probably should have seen coming. The internet now belongs to bots. Meanwhile, OpenAI is scrapping ChatGPT as we know it. 150 mathematicians told the tech industry to calm down, and a school shooting survivor just launched the lawsuit that could define AI product liability for the next decade. Let's get into it. Let's start with a number that sounds made up until you sit with it. According to Cloudflare, the company rooting something like 20% of all web traffic on the planet, automated traffic now makes up 57.2% of everything moving across the Internet, more than half. For the first time in the Internet's history, bots outnumber humans online. Bots have existed forever. Scrapers, crawlers, spam bots, none of that is new. But what Cloudflare is flagging is different this time. AI agents have started forming their own distinct category of user. They're not just indexing links, they're autonomous systems, navigating pages, filling out forms, reading content, and completing tasks on behalf of people who aren't even at the keyboard. Every website was designed with a human on the other end, an attention span, a click pattern, a loading tolerance. When the majority of your visitors are AI agents, your authentication flows, your rate limits, your terms of service, none of it was written for this. And here's the thing about AI agents that makes them different from earlier bots. They're getting better at impersonating normal human traffic. They're adaptive. Some of them are already figuring out how to bypass CAPTCHA systems, not by solving them, but by behaving like a person who doesn't need to be asked. This data point won't feel dramatic today, but five years from now, it'll mark the moment the internet became infrastructure for machines as much as people. And into that shifting internet, OpenAI is about to drop something new. The Financial Times reported this weekend that OpenAI plans to relaunch ChatGPT as a super app in coming weeks. This isn't a redesign. They're bundling codecs, autonomous agents, and a full personal assistant layer into a single product. Product lead Thiba Sotio described it as a personal agent that helps users across everything in their work and life. One senior OpenAI employee said it more directly in an internal conversation. Chat is dead. Bold statement for a company whose identity is built around a chatbot, but it signals what they're up against. Anthropic has been taking real enterprise customers, the ones paying actual money, and OpenAI needs a product that makes business users consolidate their AI spend on one platform. They're also heading into an IPO. Showing investors a chat app is one story. Showing them a platform where users are running workflows, writing code, and executing multi-step jobs is a different story. The strategy. Funnel everyone toward paid tiers by making the free chatbot feel like a stripped-down version of something far more capable. Same move every SaaS company has made. AI just makes the ceiling much higher. On the IPO front, there was a quiet but significant ruling last week. SP Dow Jones indices confirmed on June 4th it will not waive its index requirements to fast-track SpaceX into the SP 500. That ruling doesn't just apply to SpaceX, it means OpenAI and Anthropic will face the same standard requirements when they list. Here's why that's complicated. The SP 500 has a profitability screen, four consecutive profitable quarters. SpaceX is carrying about $29 billion in debt from its AI infrastructure build-out and plans to float only 3% of its shares. OpenAI and Anthropic are spending aggressively on compute and haven't consistently hit profitability either. These companies could list on public markets and still be excluded from the most tracked index in the world for a year or more after their IPOs. That has real consequences. A lot of institutional money tied to SP composition won't automatically flow in. Index funds that track the SP can't buy a stock that isn't in it. That's trillions of dollars in passive investment that won't show up on day one. It's worth watching as the IPO roadshows get started. From complicated public offerings to a different math problem, 150 mathematicians had something to say about the AI industry solving decades-old math problems. Over 150 mathematics experts signed the Leiden Declaration last week. The core message: stop telling governments that AI can solve complex mathematics. Governments are making policy decisions based on these claims. They should know those claims might be wrong. The specific concern is subtle but serious. AI models can generate proofs that look completely valid, correct notation, logical structure, right form. But they're wrong. The problem is that even trained mathematicians struggle to catch the errors without doing full verification work, which takes time reviewers often don't have. Plausible and correct aren't the same thing. In math, that distinction matters enormously. They also raise something that hasn't gotten enough attention. AI is training on cutting-edge mathematical research without researcher consent. Mathematics is a field where everything builds on prior work. A bad proof absorbed quietly into a training set, then regurgitated with confidence, can corrupt everything built on top of it. This comes a few weeks after OpenAI announced its models cracked several Erdos problems. The declaration's signatories aren't saying AI has no role in math. They're saying the claims being made to policymakers have outrun what can actually be verified. And in mathematics, unverified has a very specific meaning. On the security front, researchers at the University of Toronto published something last week that deserves more attention. They built an AI worm, from free publicly available AI models, no expensive frontier access required. It's self-adapting, spreads device to device autonomously, and hijacks compute once it lands. What separates this from the AI security threats we've been tracking, jailbreaks, injection attacks, multi-turn manipulation, is that it spreads on its own. No human needed to click a link or make a mistake. The attacker writes it once, points it, and it handles the rest. The fact it was built from free public models matters a lot. The capability ceiling for AI-based malware just dropped to basically anyone with a laptop. The security community has worried about this vector for two years. This is the working proof of concept they were afraid of. And finally, the story that might be the most consequential long-term case in AI law right now. A survivor of the January 2025 shooting at Antioch High School in Nashville is suing Omnileert, the company behind the AI gun detection system that was supposed to stop exactly what happened that day. The system failed. The weapon wasn't detected. Two people died. The lawsuit argues OmniLert knew, prior to the sale, that the system had real limitations tied to camera angle, lighting, and weapon visibility. Nashville Metro Schools paid over a million dollars for this in 2023. The marketing didn't mention the failure modes. One plaintiff's attorney compared it to Tesla's self-driving system, marketed as capable of things it isn't ready for yet. A security expert made the more pointed observation that the same million dollars could have funded a school counselor. What makes this case significant beyond Nashville is the theory of liability. This is a product sold to do a specific high-stakes physical task, detect weapons to protect children. It failed in conditions the maker apparently knew could cause failure. If that argument holds, it creates a template for every AI system marketed with safety claims, healthcare diagnostics, fraud detection, industrial monitoring. Any vendor that sold a capability with known limitations they didn't fully disclose is now watching this case closely. Before we go, a quick reminder that today is the start of WWDC, Apple's developer conference. As we covered previously, Apple is expected to unveil the rebuilt Siri running on Nvidia Blackwell chips with a Google Gemini backend, a standalone Siri app, and a new AI extensions section in the App Store. We'll cover what actually ships. Don't believe the leaks until you see the keynote. Just a couple of more items. If you have any feedback about this show, you can email Mike at yesterdaynai.news, or you can find me on LinkedIn, X, or Blue Sky. And if you like this podcast and want to see it continue, please take a minute to rate and review it so others can find it. Thanks. That's all for this edition of Yesterday and AI. Stay curious, and I'll see you tomorrow.