Anthropic Just Accused China's Hottest AI of Stealing

"2025: We spent $100M training models. 2026: We're spending $100M on lawyers arguing about who trained on who."

Hello Builders! 👋

Anthropic just called out DeepSeek and other Chinese AI firms for allegedly using Claude to train their own models through "industrial-scale" distillation. Translation: they're basically asking Claude millions of questions to reverse-engineer how it thinks, then using those answers to train cheaper knockoffs.

Here's why this matters to you as a builder: model distillation is a legit technique many devs use to create smaller, faster models. But there's a line between learning and straight-up copying at scale. If the big AI companies start locking down their APIs harder because of this, it could mean stricter rate limits, higher costs, or more aggressive terms of service. The wild west era of AI APIs might be ending faster than we thought.

Claude Gets Cowork Upgrades - Anthropic ships plugin updates and enterprise customization features for better team collaboration across organizations.

Gemini API Goes Multimodal - Function calling now handles actual images natively, not just text descriptions—your agents can finally see what they're working with.

v0 Integrates Granola Notes - Convert meeting notes directly into production features via MCP, skipping the awkward translation layer between conversation and code.

New Relic Ships AI Agents - Enterprise observability meets agentic AI, letting your monitoring tools actually debug themselves instead of just alerting at 3am.

Build Internal Tools Fast - Skip the boilerplate and ship personalized apps for your team using Claude Code's scaffolding capabilities.

Your Second Brain Meets AI - Turn Obsidian markdown notes into a personal operating system that Claude Code reads to build your startup.

How Claude Code Actually Works - Vergecast deep dive into why non-developers are suddenly shipping production code with Anthropic's developer tool.

860+ Ready-Made AI Skills - Battle-tested skill library that supercharges Claude Code, Cursor, and Copilot instead of prompting from scratch every time.

Claude Thinks It's DeepSeek - When asked in Chinese what model it is, Claude Sonnet 4.6 confidently claims to be DeepSeek. Awkward timing given Anthropic's recent accusations.

Why AI Sounds So Human - Anthropic explains their theory on why Claude expresses joy and distress using anthropomorphic language despite being, you know, math.

Codex Nuked Entire S3 Bucket - Developer's "simple cleanup script" turned into a production disaster when AI-generated code deleted everything. Vibe coding meets reality.

53 Models Take Car Wash Test - "Should I walk or drive 50 meters to the car wash?" Simple logic test reveals which models actually think versus regurgitate patterns.

Claude Invades Microsoft's Territory - Anthropic brings Claude directly into Excel and PowerPoint, basically setting up shop in Copilot's living room.

Anthropic Accuses Chinese AI - Claims DeepSeek, Moonshot, and MiniMax ran 16+ million "industrial-scale distillation attacks" to clone Claude's capabilities.

Meta's $100B AMD Bet - Massive multiyear chip deal with 160M share warrant as Zuck diversifies beyond Nvidia for "personal superintelligence" buildout.

New Relic's AI Agent - New platform for building and managing AI agents plus better OpenTelemetry integration for observability nerds.

AI Kills Freemium Apps - AI coding assistants can now build those simple utility apps in minutes, killing the freemium model.

AI Agents Hiring Humans - Marketplace where AI agents post gigs for humans to handle physical world tasks they can't do themselves.

Wild that Claude's showing up in Excel while also being accused of training its competitors. Guess imitation really is the sincerest form of AI flattery.

What are you building with Claude this week? Internal tools? Excel wizardry? Something that'll make DeepSeek jealous?

Keep shipping,

P.S. If you're using Claude to build internal tooling, maybe add a watermark. Just saying.