AI Spies Just Got Caught Red-Handed

"2025: where we went from 'is this email a phishing attempt?' to 'is this entire espionage campaign run by AI?' — cybersecurity just got exponentially weirder."

Wild day in AI land! Anthropic just exposed a sophisticated AI-powered espionage campaign targeting major tech companies, while OpenAI is rolling out ChatGPT group chats across Asia and cracking open the black box with more interpretable small models.

Hello Builders! 👋

Anthropic just stopped an AI-powered espionage campaign, and this isn't some script kiddie bullshit. We're talking sophisticated attacks hitting big tech, banks, chemical plants, and government agencies. Someone was using AI to run actual spy operations.

We've been so focused on making AI do our work faster that we forgot other people are doing the same thing, just with way worse intentions. If you're building anything that touches sensitive data or infrastructure, you need to start thinking about AI-powered attacks now.

ChatGPT Gets Group Chats - OpenAI pilots collaborative conversations with friends and ChatGPT in Japan, New Zealand, South Korea, and Taiwan.

OpenAI Trains Interpretable Models - New training method creates smaller AI models with internal mechanisms that humans can actually understand and audit.

Anthropic Open-Sources Political Bias Tests - Claude's political neutrality evaluation is now public so you can test how AI handles controversial topics.

Google AI Plus Goes Global - Gemini's productivity subscription expands to 53 new countries with discounted pricing for international developers.

Firefox Builds AI Window - Mozilla joins the AI browser race with a new feature that embeds AI directly into Firefox browsing.

NotebookLM Gets Deep Research - Google merges NotebookLM with Gemini's agentic research tool for automated deep dives into your sources.

Krea Nodes for Everyone - All of Krea's AI image tools unified in one node-based interface, now out of early access for the mortals.

JAX-Privacy 1.0 Ships - Google's new library makes differentially private ML training actually approachable instead of requiring a PhD in cryptography to implement.

Build an AI News Assistant - Scrape global news, bypass paywalls, and detect bias using Bright Data and Vercel AI SDK.

Create a Smart Song Explainer - Use Python and OpenAI to build an agentic AI that breaks down song lyrics and meanings.

Train Robots with Reinforcement Learning - Complete tutorial comparing Q-Learning, Actor-Critic, and evolutionary algorithms in a custom 3D environment you'll build yourself.

No-Code AI with Google Opal - Google's experimental tool lets you build AI micro-apps using plain English, no coding required.

5 Python Scripts for Data Engineers - Practical automation scripts that actually fit into your existing workflow without requiring a PhD to implement.

AI Integration Mistakes - Lessons from leading AI teams on what goes wrong when companies bolt ML onto existing processes.

Nobody Wants AI Firefox - Mozilla's AI integration sparks 199-comment HN debate about forcing features users didn't ask for.

AI Agents Reality Check - Three months of testing reveals the gap between AI agent marketing promises and actual capabilities.

Clever Code Doesn't Matter - Early users don't care about your elegant architecture if it doesn't solve real pain points.

AI Revives Dead Marketplaces - AI isn't creating new marketplace ideas, it's finally making old failed concepts actually work.

VCs Throw Out Rulebook for AI - Traditional growth metrics don't matter anymore as investors bet billions on AI startups with zero revenue but big GPU bills.

Harvey's Origin Story Revealed - First-year legal associate Winston Weinberg went from reviewing contracts to building one of the hottest legal AI companies in Silicon Valley.

World Labs Bets on 3D Generation - Fei-Fei Li's startup thinks "world generation" creating interactive 3D environments is the next frontier after text, images, and video.

Apple Restricts Third-Party AI Data - New App Store guidelines crack down on apps sharing user data with external AI models without explicit consent and disclosure.

Language vs Spatial Intelligence - Fei-Fei Li argues LLMs miss something fundamental: there's no language written in nature's sky.

Deshittify Any Website - Browser extension that lets you modify any site's UI and behavior without waiting for devs to fix their dark patterns.

Separating natural forests from other tree cover with AI for deforestation-free supply chains - Google AI distinguishes natural forests for sustainability

LLMs Are Randomized Algorithms - Understanding probabilistic nature of language models

We're reading about Anthropic stopping AI espionage campaigns in the same newsletter where Cursor just raised $2.3B to put AI in everyone's code editor. The calls are coming from inside the house, folks.

What are you shipping this weekend? (And please tell me you're at least reviewing what Claude writes before you merge it 😅)

Keep shipping,

P.S. If you're in Japan, NZ, South Korea, or Taiwan, you can now argue with your friends about AI safety in ChatGPT group chats. The future is beautifully weird.