Welcome to Jumble, your go-to source for AI news updates. This week, Anthropic accused Alibaba of the largest distillation attack it has ever seen. Days later, the White House told OpenAI to hold its newest model back from the public. Let’s dive in ⬇️
In today’s newsletter:
🪪 Anthropic gets got by Alibaba
🐢 Washington wants OpenAI's next model kept on a leash
🖱️ Gemini can now run your screen by itself
🛣️ Amazon's driverless toaster gets a glow up
📚 Weekly Challenge: Turn your notes into a personal tutor
🔍 Anthropic Accuses Alibaba of Copying Claude at Scale
Anthropic told Congress that Alibaba ran the largest distillation attack it has ever seen, quietly milking Claude's smartest abilities through thousands of fake accounts.
Is copying a rival's AI theft or fair game?
🧪 What Actually Happened
Operators tied to Alibaba's Qwen lab allegedly used roughly 25,000 fraudulent accounts to run 28.8 million exchanges with Claude between April and June. They went straight for its best coding and reasoning skills.
🛡️ Why It Matters
Distillation lets a rival rebuild a top model on the cheap, often stripping out the safety guardrails baked into the original. Anthropic warns that makes the copies riskier, not just cheaper.
🏛️ The Washington Angle
The accusation lands weeks after the White House forced Anthropic to cut foreign access to its top Fable and Mythos models. A massive Chinese copying campaign makes that clampdown look far less paranoid.
🛑 The White House Tells OpenAI to Slow Down GPT-5.6
Days after restricting Anthropic, Washington turned to OpenAI, asking it to hold its newest model, GPT-5.6, back from the public and release it only to a small set of approved partners first.
|
🥇 A First for Washington
It is the first time the government has preemptively asked a US lab to limit a model's launch. The request came from the White House cyber and science offices, not the agency that hit Anthropic.
However, just a few weeks ago, we were all warned that this was the future, so it shouldn’t come as much of a surprise.
⚡ Why GPT-5.6 Spooked Them
Officials and OpenAI reportedly see it as on par with Anthropic's Mythos, with cyber skills that could find and exploit software flaws at inhuman speed. With no clear rulebook for who polices frontier AI, OpenAI is quietly doing what Anthropic already chose to do.
Weekly Scoop 🍦
👩🏫 Weekly Challenge: Turn Your Notes Into a Personal Tutor
Challenge: Google just launched study notebooks in the Gemini app, a free space that turns your own notes into a custom study plan. This week, point it at something you are trying to learn.
Here's what to do:
📒 Step 1: Start a study notebook. On the web at gemini.google, open the side panel, pick "New notebook," then "Study," and set your topic or goal.
📤 Step 2: Feed it your material. Upload your notes, slides, a syllabus, or readings, then take the quick diagnostic quiz so it can spot your weak areas.
🎯 Step 3: Work the plan. Run through the bite sized lessons and follow up quizzes, and watch the dashboard track what you have actually nailed.
If a rival can copy a frontier model in weeks, is locking the best ones away the only real defense? And do you trust Washington to decide when a model is safe to ship? See you next time! 🚀
Stay informed, stay curious, and stay ahead with Jumble!
Zoe from Jumble
