Week 27: AI Spreads Hate, Web Locks Down, Zuck Finds Religion
AI IS NO LONGER A NOTE TAKER
Every week AI learns something new, and every week we learn we should have seen it coming. This time: Google's video AI became a hate factory, vision systems learned to hallucinate on command, and Mark Zuckerberg found his next expensive obsession.
HUMANS FIGHT BACK
Web Slams Door as Cloudflare shifts its policy to make AI bot scraping opt-in for customers, meaning content will now be blocked from AI by default.
AI BREAKS BAD
Hate Scales Industrial as Google's Veo 3 AI video generator is reportedly used to create and spread racist and antisemitic content on TikTok, garnering millions of views. Vision Gets Hacked as researchers developed "RisingAttacK," a method to manipulate AI vision systems, making them "see" whatever the attacker desires with minimal image changes.
MONEY CHANGES EVERYTHING
Cursor Prints Cash as Cursor, an AI-powered developer tool, reached $500M ARR in just 30 months by leveraging OpenAI's scaling laws. Zuck Changes Religions with Mark Zuckerberg intensely promoting a new "AI superintelligence" vision, drawing parallels to his previous, largely unsuccessful, metaverse pivot.
Curious what it all adds up to? Let’s break it down. Keep reading below.
Tell Me More
The Free Lunch Ends. Cloudflare's new default: block AI scrapers unless companies pay up. This isn't just about data access — it's about who controls the raw materials of intelligence. Small creators might soon earn more from AI licensing than from human readers, completely flipping the economics of online publishing. By 2027, expect a two-tiered internet: premium AI-accessible content and a "commons" of lower-quality free data. Even robots will need premium memberships. mediacopilot.substack.com
The Hate Machine Scales. Google's Veo 3 churns out antisemitic content faster than human moderators can blink. The crisis is forcing an uncomfortable question: what happens when creating hate becomes cheaper than stopping it? Within 18 months, platforms will implement "proof of human" verification for all uploads. The unexpected outcome? This might actually reduce overall hate speech by making human-generated bigotry easier to identify and prosecute. The internet is splitting into authenticated human spaces and AI-generated wastelands. arstechnica.com
Vision Under Attack. With the right pixel tweaks, researchers can make AI "see" elephants as mice or stop signs as speed limits. The implications for autonomous vehicles are terrifying. These vulnerabilities might actually accelerate AI adoption in some fields. When humans know the system can be fooled, they demand multiple verification layers, making AI-assisted decisions more rigorous than human-only ones. AI's flaws are becoming features that force better system design. interestingengineering.com
Zuck's Pivot Playbook. Mark Zuckerberg collects ambitious visions like vintage cars — beautiful, expensive, and often left to rust in the garage. His leap from metaverse to AI superintelligence has all the hallmarks of a rich kid's attention span: breathless excitement, blank checks, and the lingering smell of abandoned dreams. Here's the thing: Zuckerberg's pattern of expensive pivots might actually be optimal strategy — in a world of exponential change, the cost of being wrong is lower than the cost of being late. Silicon Valley has evolved from "move fast and break things" to "move fast and abandon things." arstechnica.com
Below The Fold
A "10x engineer" allegedly scammed multiple AI startups by working multiple full-time jobs simultaneously, earning over $1M annually while under-delivering across all positions. Pragmatic Engineer, The Verge
Surveillance app Catwatchful leaked 62,000 users' plain-text passwords, proving that watching the watchers is harder than watching everyone else. Ars Technica
Wilco’s Jeff Tweedy uses timed musical improvisation to "recalibrate" after travel, turning jet lag into jazz therapy. Jeff Tweedy
AI note-takers have become the ultimate meeting escape hatch, letting workers ghostwrite their own attendance. Washington Post
People are hiring AI chatbots as digital shamans, proving that even consciousness expansion can be outsourced to algorithms. MIT Technology Review
A tech veteran mourns the death of the "creative-tech era," when individual expression mattered more than algorithmic optimization. greg.technology
A Melbourne man discovered an elaborate model train network, complete with tunnels and tracks, hidden beneath his newly purchased home. SBS
The White House's favorite news source about itself is... itself, creating the ultimate media echo chamber. The Verge
A new open-source project aims to provide a self-hostable continuation of the IRS Direct File program, pushing for digital public goods. GitHub
The most valuable skill in AI isn't prompting, but "context engineering," structuring information for meaningful results. Phil Schmid
Looking Ahead: As AI systems demonstrate increasing agency and power, next week's developments may force us to define the true boundaries of human control.
Thanks for reading Briefs — your weekly recap of the signals I couldn't ignore. This week that meant reading 632 stories from 45 sources. You're welcome.