💎 The RubyGems attack and Claude token theft

⚡ aiBytes_ #9

Hello builders 👋

Sachin here! Every week, I spend a few hours exploring what’s happening in AI and picking the resources worth sharing with you. This newsletter feeds my own curiosity, too. There’s a lot to keep up with, so I put the most useful finds in one place. If it helps you, please forward it to a friend.

Today’s picks include the RubyGems investigation, DeepSeek v4.1 Flash, Mastra Factory’s delivery workflow, archify’s architecture diagrams, and a tiny challenge to keep Claude focused on making one button blue.

The RubyGems story is the one I’m still thinking about. Researchers link a May attack to OpenAI agents. Their findings leave open questions, but they give the discussion about agent safety something concrete to examine.

💎 Researchers link the RubyGems attack to OpenAI agents - The analysis traces malicious packages to a May incident; authors say the agents’ motives and whether key theft succeeded remain unknown. Story of the week (RubyGems investigation)

🧮 NYU mathematician disputes OpenAI’s conduct - A dispute over credit surrounds the Navier-Stokes problem. (TechCrunch)

⚡ DeepSeek announces v4.1 Flash - The new model announcement drew a busy HN discussion; read the original release before choosing what to test. (DeepSeek)

🇫🇷 Mistral raises €3B - New funding backs its push for sovereign, open-weight AI. (Mistral)

🔑 Claude subscribers report stolen token usage - Anthropic has warned users about hackers after unexplained consumption appeared on accounts. (TechCrunch)

🧑‍🔬 Anthropic researcher resigns over self-improving AI - Jacob Coxon calls for pacing agreements between labs, according to TechCrunch. (TechCrunch)