Anthropic Researcher Quits, Says AI Labs Are Gambling With Our Lives!

Today's AI news: pretraining researcher Jacob Coxon resigned from Anthropic saying neither OpenAI nor Anthropic is acting responsibly and both are racing to self-improving superintelligence. Anthropic's Economics team published a model of AI's effect on US growth, jobs and wages by 2030. Claude Marketplace added CrowdStrike, Cursor, Factory, Gamma and Vercel. Unity shipped an official first-party plugin for Claude Code. And Google Research and HHMI Janelia released the complete fruit fly connectome, which developers promptly got playing Doom. Chapters: 0:00 Today's AI News 0:24 Anthropic Resignation 1:13 Anthropic Economics Model 1:51 Claude Marketplace 2:30 Unity Claude Code Plugin 3:13 Fly Brain Connectome Blend Roundup 2026-09-09 https://x.com/hilbertspaess/status/2097476196791709843 https://x.com/AnthropicAI/status/2097679796687769689 https://x.com/claudeai/status/2097718980437831935 https://x.com/unitygames/status/2097722273696227338 https://x.com/nftechie_/status/2097711063135383642 https://x.com/GoogleAI/status/2097678989695324609 An Anthropic researcher has quit: saying the labs are gambling with our lives. Anthropic models the AI hit on jobs by 2030, Claude Marketplace adds Cursor and CrowdStrike, Unity ships an official Claude Code plugin, and a fruit fly's brain is playing Doom. Here's today's AI news. Jacob Coxon, three years a pretraining researcher at OpenAI and then Anthropic, resigned yesterday with a post that's now been seen well over a hundred million times. His claim is that neither company is acting responsibly: both are racing straight to self-improving superintelligence and, in his words, gambling with our lives. He told the Wall Street Journal that things could be out of control by the end of next year, and put the chance of AI killing everyone above ten percent. His read on the two labs is the striking part: at OpenAI, many staff haven't internalised the stakes; at Anthropic they have, but they believe nobody else will act responsibly, so they race to get there first. Anthropic's alignment lead publicly confirmed the company earnestly believes AI could kill everyone. This confirmation is the real headline. Anthropic's Economics team has published a model of what AI could do to the US economy by 2030, with an interactive explorer and a survey of nearly eleven thousand Americans. It breaks jobs into tasks and asks whether AI speeds each one up; does it outright; leaves it alone or creates new ones. Three scenarios come out: modest, where GDP is under two percent higher and unemployment barely moves; substantial; where AI can do half of all knowledge work, growth doubles and knowledge-worker wages stall; and extreme; with fifteen percent annual growth and knowledge-worker unemployment near eighteen percent. Anthropic has added CrowdStrike, Cursor, Factory, Gamma and Vercel to Claude Marketplace. Enterprises that have already committed spend with Anthropic can now put that money towards third-party Claude-powered tools without opening a new procurement cycle, with Anthropic handling the invoicing. So a CrowdStrike Falcon integration, a Cursor licence or Vercel deployment becomes a line-item reallocation rather than a fresh contract. It's in limited preview. The interesting bit is who benefits: partners get access to budgets that were already signed, and Anthropic gets to sit in the middle of every one of those deals. Quietly, that's a platform. Unity has shipped an official, first-party plugin for Claude Code, available straight from Anthropic's plugin directory. One install, no configuration: it drops in twenty nine native Unity skills at launch - covering UI, 2D and tilemaps, in-app purchases, localisation, multiplayer and Shader Graph - plus the Unity CLI and an MCP server for live control of the Editor, so Claude can work inside a Unity project the way Unity intends rather than the way it guesses. Claude Code is first; Unity says other coding agents get the plugin later this year. Games created with Astra have been wowing everyone; let's see what Claude and Unity can produce. Google Research and HHMI Janelia have released the complete wiring diagram of a male fruit fly's brain and central nervous system - a hundred and sixty six thousand neurons and a hundred and twenty five million connections, the largest proofread brain map ever made, after a decade of work. Within days, developers had turned the open dataset into a rough simulation and got it playing Doom, then Super Mario 64, then Beat Saber. It's a static anatomical snapshot with heavy approximations, not a living brain, so nobody should read too much into the frag count. But a complete nervous system you can download and poke at is a genuine milestone.