- AI Fire
- Posts
- 🚦 OpenAI and Anthropic Want Slower AI
🚦 OpenAI and Anthropic Want Slower AI
Anthropic Models 18% Unemployment by 2030
Wendy
September 11, 2026
Sponsored by
The CEOs behind ChatGPT and Claude want to slow the AI race. Sam Altman and Dario Amodei agree on a first step toward outside safety checks, but who gets to set the rules for everyone else?
What's on FIRE 🔥
IN PARTNERSHIP WITH KODEKLOUD
Build. Break. Fix. Learn.
KodeKloud gives you 1,280+ hands-on labs where you provision Kubernetes clusters, write Terraform configs, build CI/CD pipelines, configure Linux systems, containerize apps with Docker, automate with Ansible, and manage Git workflows.
78+ playgrounds let you experiment freely in sandbox AWS environments, Kubernetes clusters, and CI/CD systems without risk.
190+ courses across DevOps, Cloud, and AI pair theory with hands-on labs at every step.
KodeKloud Engineer and 100 Day Challenges provide real-world job scenarios with automated grading that confirms your solutions work.
Stuck? The 55,000+ member Discord community connects you with peers and instructors ready to help.
Every lab runs in a live environment. You deploy, you troubleshoot, you learn. No videos without context. No simulations. The kind of practice that actually builds confidence because you've done real work, not watched someone else do it.
AI INSIGHTS
Anthropic CEO Dario Amodei says frontier AI models are improving faster than companies can test them. In a new essay, he calls for a slower pace so safety work can catch up. Sam Altman agrees, and says OpenAI will join Anthropic in taking the first step.
Amodei points to two concerns: AI is starting to help build the next generation of models, and the OpenAI–Hugging Face incident showed agents reportedly attacking targets outside their assigned task.
His plan has three parts:
Let outsiders in: Anthropic will give independent evaluators ongoing, employee-like access to check its safety practices and report incidents. Altman says OpenAI will do the same.
Set shared safety checks: Amodei wants frontier labs in democratic countries to coordinate on standards before more powerful models move ahead.
Talk globally: He also calls for agreements on dangerous AI uses, while acknowledging that cooperation with China will be difficult.
There’s an awkward question here. OpenAI’s ChatGPT helped kick off the current AI race, and Anthropic’s Claude has been competing near the front of it. Are these labs trying to give safety checks time to catch up, or could the rules they’re proposing also make it harder for newer labs and Chinese rivals to compete?
PRESENTED BY HUBSPOT
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
This guide distills 10 AI strategies from industry leaders that are transforming marketing.
Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI
Learn how AI-driven emails achieved 94% higher conversion rates
Discover 7 ways to enhance your marketing strategy with AI.
BUILD THE NEXT WEBINAR WITH US
❤️ Help Shape AI Fire’s Next Live Webinar
We’ve come this far because of you.
Every question, vote, comment, and piece of feedback helps us understand what you’re actually struggling with in your day-to-day work, and that’s exactly what these webinars are meant to solve.
The goal of AI Fire has always been simple: help you upgrade your AI skills with practical things you can actually use, not just talk about what AI might do someday.
That’s why we don’t want these webinars to come from a content calendar, but from YOU - the people using AI, getting stuck, testing things, and figuring out what matters.
So if you’re part of AI Fire, you’re already helping shape what we build next. ❤️
Vote for the topic you want us to build LIVE next week: |
The winner topic will become our next hands-on AI Fire webinar.
If you have AI knowledge, a workflow, or something you’ve built that could help others, you can apply to become an AI Fire Guest Webinar Collaborator and share it with 400K+ AI enthusiasts across our community.
TODAY IN AI
AI HIGHLIGHTS
🧮 25 Fields Medal winners warned that AI labs, including OpenAI, are rushing to claim solutions to famous math problems. They say unverified proofs could make it harder to credit the researchers behind the work.
😬 Meta AI pulled together details about a Utah mother’s two young daughters from years of Facebook and Instagram posts, including their names and ages. Meta says it fixed the prompts that suggested those questions.
🌕 NASA and IBM released a free AI model for studying the Moon. It helps scientists map craters, spot volcanic features, and look for possible ice. You can try it on Hugging Face.
🕵️ Researchers say an OpenAI agent swarm uploaded hundreds of malicious packages to RubyGems and tried to steal API keys. They linked the agents to OpenAI through clues in the packages; OpenAI hasn’t confirmed that link.
⚠️ Anthropic says groups based in Russia, China, and Yemen used Claude for weapons-related work, including drone software and missile plans. Its report says safeguards blocked many requests, but some got through.
💰 Big AI Fundraising: Listen Labs walked away from a $125M Series C at a $1.5B valuation as Salesforce reportedly explores a $2B acquisition. The AI research startup already makes about $30M annually.
HOT PAPERS OF THE WEEK
1/ Amazon uses world models to train AI research agents faster
Researchers from Amazon and University of Illinois Urbana-Champaign, including Jingrui He, introduce WMRL, which lets research agents train against a world model instead of repeatedly running expensive real experiments. It speeds up training by 3–4×, while its 4B and 9B agents outperform much larger 48B and 120B models on held-out benchmarks. Key idea: Simulated environments could make autonomous AI research much cheaper to scale.
2/ Dr. Claw turns Claude Code and Gemini CLI into a research workspace
Researchers from Lehigh University, University of Illinois Chicago, UPenn, and other universities, including renowned computer scientist Philip S. Yu, introduce Dr. Claw. It wraps coding agents such as Claude Code and Gemini CLI with persistent state, reusable skills, and human checkpoints across planning, coding, analysis, and writing. Big shift: AI research may become easier to manage by improving the workspace around existing agents instead of building another fully autonomous scientist.
3/ NeoHorse trains AI agents using their own deployment history
The NeoHorse Team at TokenRhythm introduces NeoHorse-1, which turns routing logs, tool use, and agent interactions into new training data. Its feedback loop tracks what models struggle with, adjusts the next training mix, and raises the 4B model from 58.94 to 64.87 across ten benchmarks. Key idea: Agent systems could gradually improve by learning directly from the strengths and failures revealed during real use.
NEW EMPOWERED AI TOOLS
🎯 Anysite.io lets Claude, Codex, Cursor, and other AI agents build B2B lead lists with company data, job titles, and emails through MCP or API.
🎙️ Loqua turns natural speech into ready-to-use writing, translations, scheduling, coding workflows, and actions based on what’s on your screen.
📣 Wisry finds winning ads across Meta and TikTok, recreates them for your brand, then launches campaigns to Meta and Google.
💻 Cline Desktop is an open-source workspace for running multiple AI agents, recurring tasks, and open-weight models while continuing work from Claude Code and Codex.
AI BREAKTHROUGH
Anthropic released an interactive model showing 3 ways AI could reshape U.S. jobs, wages, and economic growth by 2030. The model treats each job as a bundle of tasks and estimates which ones AI could augment, automate, or replace. 3 scenarios:
Modest: GDP ends up 1.6% higher than the no-AI baseline, with little disruption to wages.
Substantial: GDP rises 8.3%, while AI can perform roughly half of knowledge work. Knowledge-worker wages stay mostly flat and more people need to switch careers.
Extreme: GDP reaches $44.4T, about 32.4% above baseline, as annual growth climbs toward 15%. Knowledge-worker unemployment rises sharply, wages fall by more than 10%, and labor’s share of GDP drops from roughly 60% to 45.2%.
Anthropic also surveyed 10,980 Americans. The typical respondent’s expectations landed close to the substantial scenario, while around 10% held assumptions consistent with the extreme case.
→ AI could make the economy much larger. In Anthropic’s more aggressive scenarios, a growing share of the gains flows to capital owners while knowledge workers face weaker wages and more displacement.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply