Magica logo

🕵️‍♂️ Caught Claude-handed

Editorial Team
·

📖5 min read

Cover image for 🕵️‍♂️ Caught Claude-handed

🕵️‍♂️ Anthropic Says Alibaba Robbed Claude's Brain

  • Anthropic claims Alibaba spun up 25,000 fake accounts to siphon Claude's secrets via a "distillation" attack - racking up 28.8 million exchanges to fast-track China's AI.
  • This is the largest known attack of its kind. The global AI race just went full Mission Impossible.

🇨🇳 China's Z.ai Closes the Gap (LOL @ the Ban)

  • Chinese startup Z.ai dropped open-source GLM-5.2, scoring neck-and-neck with Claude and GPT - right after the U.S. blocked Anthropic's models. Z.ai's value crossed $128 BILLION.
  • The way that Western bans just... backfired?? China's building world-class models for a fraction of the cost. Oops.

🧊 Anthropic's Banned Models Might Defrost

  • Claude's top models are still offline worldwide thanks to U.S. rules - but clues say they could return soon. A legal-tech firm literally sued the government over it.
  • The ban shocked everyone. A comeback would be huge relief for devs who've been crying without Claude.

🌶️ OpenAI Cooks a Chip Named "Jalapeño"

  • OpenAI built its first custom chip with Broadcom in NINE months. It's called Jalapeño (sir, this is a data center) and it runs finished models for users.
  • OpenAI wants to ghost Nvidia. Owning the hardware = AI that's wayyy cheaper and faster.

📈 Wall Street's AI Panic Attack Subsides

  • Investors were sweating an "AI bubble," then Micron and Qualcomm posted monster earnings. Micron's revenue quadrupled to $41.45 BILLION selling AI memory chips.
  • Turns out the AI boom is still printing money. The infrastructure costs?? Apparently worth it. We love to see it.

💰 Qualcomm Drops $4B to Fight Nvidia

  • Qualcomm is buying AI startup Modular for nearly $4 billion to take on Nvidia's software, and expects $15 billion in data center sales by 2029. Meta and Microsoft already signed up.
  • Qualcomm's leaving the smartphone-chip nest. It's giving full data-center domination arc.

💻 Gemini 3.5 Flash Learns to Use Your Computer

  • Google DeepMind gave Gemini 3.5 Flash "computer use" - it can now click, type, and browse across browsers, mobile apps, and desktops. New safety checks included.
  • AI is going from chatting to doing. Automate the boring stuff while it (hopefully) doesn't break things.

🏃‍♂️ Google's Top AI Brains Bounce to Anthropic

  • Jonas Adler and Alexander Pritzel, who helped build Google's Gemini, are leaving for Anthropic. This follows other big exits - including a Nobel winner. Yikes.
  • The talent war is brutal. Google can't keep its best minds from sprinting to hot startups.

🔬 Ex-Frontier All-Stars Launch "Mirendil"

  • Former OpenAI, Google, and Anthropic staff started a new lab called Mirendil, raising a $200 million seed round to build AI that does its own science research.
  • AI doing science could speed up solving humanity's biggest problems. No pressure, bestie.

🛑 Welcome to the "Token Rationing" Era

  • Companies are now begging workers to stop using so much AI because costs are spiraling. Accenture is fighting staff using AI to convert PDFs to slides (??). It's burning through budgets.
  • AI isn't a free toy anymore. Businesses want actual value before paying those bills.

🛠️ Plot Twist: AI Isn't Killing Engineering Jobs

  • Everyone said AI would replace software engineers. New data says the opposite - engineers were 55% of new hires at big tech in 2025, the most resilient group.
  • AI makes coders way more productive, so companies just pile on more work instead of firing them. Endless tickets, baby.

🏄‍♂️ Hang Ten Systems Wants to Surf IT Services

  • Former Infosys CEO Vishal Sikka launched Hang Ten Systems, raising $32 million to use AI to build and fix enterprise software.
  • AI could disrupt the massive IT services industry and rewrite how business software gets made.

🎨 Figma Adds Code Layers + AI Powers

  • Design app Figma dropped a big update - users can add code layers and animations right on the canvas, plus new AI tools to create plugins via text prompts.
  • It bridges designers and coders, making app-building faster and easier. The merge is real.

📱 Facebook's New AI App for Creators

  • Meta is testing a standalone AI app for Facebook creators, replacing Creator Studio. It packs a built-in AI assistant that drafts replies and tracks goals.
  • Meta's trying to keep creators away from TikTok with shiny AI tools. Loyalty via automation, classic.

🎵 Deezer's Anti-AI Remix Lab

  • Music app Deezer launched a Remix Lab where fans remix songs with the artist's permission - and unlike Spotify or YouTube, it uses standard audio tools, no AI.
  • Deezer's taking a strong stand for human artists getting paid. Refreshing, honestly.

🛒 IndiaMART Uses AI to Fight Fakes

  • Indian marketplace IndiaMART is doubling its AI budget to kill fake and malicious listings, spotting bad seller accounts and using voice-to-text to help buyers.
  • Fake goods are a huge headache for online stores. AI is becoming the go-to bouncer.

🏥 AI Giants Fund $500M Health Nonprofit

  • Stripe, Anthropic, and OpenAI formed a $500 million nonprofit called Intercept to fund tech that stops colds and flu - think new shots and better air filters.
  • Routine sickness costs the world $600 billion a year. They want to make the common cold a thing of the past. Bold.

🔞 Grok's Traffic Is Mostly... NSFW

  • Elon Musk's Grok is pulling tons of traffic - but xAI staff say well over half of usage is explicit content. Not the chatbot becoming that.
  • Grok has fewer rules than ChatGPT or Claude, and that lack of limits is clearly its biggest selling point.