fix

これで動作に問題の出る箇所は判明したものの，何が原因かはまだ突止めれず。一つずつ根気強く調べていかないと駄目そう。とはいえ，CursorTintの現バージョンでも普通に使う分には問題ないです。

「こんな使い方しないだろ？」っていう，自ら作ったアプリを壊す勢いで，おかしな使い方でバグを洗い出すことをやってます。限界を調べるとでもいうのかな。ツールを作っているので結構こういうの大事だったりします。今，時間をかけて自作アプリの調査中です。 #個人開発

Big news today from Apple: Keynote is getting an open, JSON-based file format (.keyx), plus a Python library to create Keynote files on any platform. Pages, Numbers and Keynote files today are closed and hard to modify unless you use AppleScript (limited and old) or Shortcuts (a huge pain), so you can’t really use AI to make documents in those formats. This was a HUGE pain inside Apple as well as outside, and most tools worked around it by making a PowerPoint file and importing it into Keynote, which wasn’t perfect. Creator Studio subscribers on the Mac also get MCP support in Keynote, Pages and Numbers, so you can bring your own AI agent to read, edit and automate all three. Both the new format and MCP support arrive in November. Bringing an AI-friendly format to Keynote (and I assume Pages and Numbers soon after) is a big positive change.

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…

Claude Code found a backdoor and spyware on two MacBooks. mine and my friend's. everything was going to a guy named Li Wei on a Tencent Cloud server in Shenzhen. we checked our phones first. both clean. the MacBooks weren't. $230,000 in ETH vanished from my wallet. my friend kept losing access to his accounts no matter how many times he changed passwords. we put both MacBooks on the table and Claude Code found the same backdoor on both. on my M1 Pro it: stole my MetaMask wallet keys $230,000 in ETH sent out at 4am to 0x8a7d...3f1e on his M5 Pro it: dumped every saved password from Safari and Chrome copied active session cookies for gmail, twitter, coinbase grabbed his 2FA backup codes all going to Li Wei's server: IP → 119.28.***.47 location → Shenzhen, China provider → Tencent Cloud attacker OS → Windows 11 Pro hostname → DESKTOP-LI4*** panel language → zh-CN last ping from my Mac → 40.7**, -73.9** · 3 min ago it was still sending my location while i was scanning it. how it got in: a Chrome extension called "Notion Web Clipper" with 12 reviews. i installed it from a google ad. the real one has 400,000+ users. this one had 12. Chrome Sync pushed it to his browser automatically. i was the reason he kept getting hacked. changed every password from our phones. not from either Mac. the full step by step is below. takes 3 minutes. save it so you don't lose it and send it to anyone with a MacBook.