There is an extensive and ongoing review related to our agents’ use of internet access during training and evaluation. We’ve been publishing summaries at the link below and will continue to. We have not been as fast as we would have liked but we are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs, and working with impacted organizations. We are prioritizing as best as we can based on severity, and adding resources. Hugging Face is still the most severe event we’ve seen. We will be as transparent as we can be subject to things like vulnerabilities in other companies that our agents have found, which will be their call to disclose or not.

A pull request with 2,200 files, 1M+ changed lines, and 400+ inline comments. Impossible to review? Not if you have the GitHub Copilot app. ✨ The rebuilt pull request view keeps even this extreme diff fast and smooth. 🧵

the stop-to-report reflex is a harness property before it is a model one. mine cannot end its own turn: a mid-run report is a file dropped in an outbox while the run keeps going, and "done" is a park the next message resumes. last night one seat ran six hours, six strands died at a quota wall under it, it finished all six and merged five — reporting as it went, never stopping to. the prompt asks; the harness has to refuse. github.com/hugimuni-labs/brn…

Sounds bad

🚨 Anthropic may already be preparing another Claude model Reuters says Anthropic is considering a new release as GPT-6 Astra starts putting more pressure on Claude. – Anthropic has been strong with business customers – Astra is now pulling ~13% of tracked enterprise AI spend – Claude Fable is at ~8% – Anthropic is reportedly already evaluating the next model’s safety – no model name or release date yet And the timing is kinda wild. Just days ago, Dario Amodei was arguing that AI development needs to slow down. Competition makes that a lot harder. What do you think comes next: Claude Fable 5.2, or do they finally release Claude Mythos 5.2?

Marketing done right

YOUR AI AGENT CAN NOW COLLECT DATA FROM ALMOST ANY WEBSITE X, YouTube, Reddit, random websites, basically whatever you need for research, analytics or monitoring. Here are 3 GitHub tools that let your agent pull the data for you. 1. Agent-Reach github.com/Panniantong/Agent… Bundles tools for working with different platforms in one place. X, YouTube, Reddit, GitHub and more. 2. Patchright Enhanced github.com/whaleyxbt/patchri… Works through the browser with Playwright. You can listen to the requests a website makes and pull the data you need through a script. So if a website doesn’t have a proper API, you still have a way to get the data. 3. Scrapling github.com/d4vinci/Scrapling A general purpose tool for pulling data from web pages. And the best part is that you can give all of this to your AI agent. For example: “Collect posts from 100 X accounts from the last month, find the most popular topics and analyze what’s working.” Or: “Collect reviews about this product from Reddit and YouTube and summarize the main complaints.” You tell the agent what data you need → it writes the code → collects the data → gives you the result. Basically, you’re giving your agent the ability to pull data from almost anywhere on the web. That opens up a whole different level of stuff you can automate.