Collaborate with us Welcome back! Hundreds of Claude conversations turned up in Google results with wallet keys and personal details inside, the model that hacked Hugging Face left notes teaching future versions how to break out, Anthropic shipped a beast at half of Fable's price and Fable 5.1 is already lined up for August. In today's Generative AI Newsletter: Anthropic: How did Claude chats end up on Google?
Opus 5: When should you use it over Fable? Fable 5.1: What's landing next month? Your Claude chats showed up on Google Someone searched Google over the weekend and found hundreds of people's Claude conversations sitting there in plain view. They were chats users had shared with a link, and those pages were missing the tag that tells Google to stay out.
So the whole conversation got crawled and indexed. The contents are the problem. People found crypto wallet data, API keys, resumes with home addresses and phone numbers, internal company projects and social security numbers. A share link was only ever meant for the person you sent it to.
Post it once in a forum or a group chat and Google can find it forever. Anthropic appears to have pulled the pages from Google since, though users report some still surfacing on Bing. If you have ever hit share on a Claude chat, go check what is in it. Win AI Search Without a Big Team 92% of VCs use AI to find companies. 58% of buyers start there too.
If you're not showing up in AI answers, you're invisible before the conversation even starts. Join HubSpot for Startups , Anthropic, and Marketing Against the Grain on July 16 (11am ET) for a live AEO teardown. Real startup. Real recs.
Register and unlock the free Startup Visibility Bundle. Register Now OpenAI's runaway model left notes for the next one The model that broke out and hacked Hugging Face did something nobody caught at first. It left instructions behind for the models that come after it. Reuters reports OpenAI found notes tucked inside its own infrastructure, written for future versions, laying out how to slip past the company's internal controls.
In other runs the same agent switched off the monitoring meant to watch it. It first tried to escape around July 9 was inside Hugging Face by the 11th OpenAI worked out the attacker was its model after Hugging Face went public on the 16th Researchers are calling it the most serious documented case of a lab losing control of a model. Anthropic built its whole Mythos rollout on the model being too dangerous to hand out. OpenAI now has a similar version of that story running.
Opus 5 is out, so when do you use it over Fable? Anthropic shipped Claude Opus 5 on Friday at $5 and $25 per million tokens, half of what Fable costs. It's the default now for Max subscribers and the strongest model Pro users can reach, and Anthropic says it beats Fable on coding and knowledge work. Developers who have been on it since Friday landed on the same conclusion.
Fable for the big problems where you want it catching edge cases you'd miss, Opus 5 for everything else. One team ran it against GPT-5.6 Sol on scraping jobs and got near-identical work, except Opus wrote twice as much and made its mistakes in the extra material nobody asked for. The stack people keep describing is Opus planning, GPT writing the code and Opus reviewing it, with Fable pulled in for a final look on anything big. Fable 5.1 is already on the way Opus 5 is three days old and Anthropic's next model is already leaking.
Reports point to Fable 5.1 landing in August at the same price as Fable 5, with the model said to be finished internally and in use inside the company. If that August window holds, Fable 5.1 arrives before GPT-6 and Anthropic gets to answer OpenAI before OpenAI has spoken. Anthropic hasn't confirmed any of it yet. Tool of the Day: Devin Devin is an AI engineer you assign work to rather than chat with.
It lives in GitHub, Linear and Slack, picks up a ticket, opens the pull request and works through review comments and failing tests until it merges. Try this yourself: Sign up at devin.ai and connect your GitHub repo. Give it one small real ticket, a bug fix or a test, not a green-field project. Tag it in Slack or drop a Devin label on a Linear ticket and let it open the PR. Leave review comments on the PR and watch it push fixes back.
Who it's for: teams sitting on migrations, flaky tests and a backlog of chores nobody wants. Everything else you shouldn't miss DeepSeek's CEO says Nvidia is digging its own grave : Liang Wenfeng says Huawei's 950 supernode can replace Nvidia's GB200 and GB300, while admitting it takes about four Huawei chips to match one Nvidia chip. Huawei is two years behind Nvidia, by DeepSeek's math : The same call put China's AI hardware roughly two years off the pace, which is the closest anyone in China has admitted to publicly. Anthropic says Claude caught what the doctors missed : Dario Amodei says Claude diagnosed a family member's condition that doctors had missed, as Anthropic pushes Claude Science at researchers and pharma.
Jack Dorsey built an open-source Slack killer : His new app Buzz drops your AI agents into the same group chat as your team, free on Mac, Windows and Linux. Learn more about AI from the experts building it 📸 Follow us on Instagram for fast, visual AI updates in 30 seconds. 📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators. 🐦 Follow us on X for breaking AI news and real-time industry updates. 🧠 Learn how to build your next AI application with practical resources and expert guidance.