Amodei Calls for AI Slowdown — The Internet Responds The Verge / Multiple outlets Anthropic CEO Dario Amodei published an essay urging every frontier AI lab to pause or decelerate development, setting off the biggest public debate the industry has seen this cycle. By Monday night the CEOs of Microsoft, OpenAI, and xAI had each responded, a US senator had proposed banning superintelligence outright, and President Trump praised Amodei on Truth Social. A draft “Humanist Superintelligence” Code of Conduct circulated alongside the debate, proposing principles including human control, safety over task completion, and a prohibition on independent AI goals.

🔥 AI Policy & Safety

Mark Zuckerberg Takes Aim at Anthropic in Debate Over AI Slowdown — The New York Times Zuckerberg pushed back directly, arguing that industrywide pauses are the wrong approach and that each lab already has its own incentive to move safely. The comments land as Meta accelerates to close the gap with Anthropic and OpenAI after falling behind last year and plans a major model release.

A Brief History of AI Executives Calling for Regulation — The Verge A pointed retrospective charting how the same executives who have called most loudly for safety guardrails — Altman, Amodei, Hassabis, Nadella — have simultaneously shipped at full pace. The piece asks whether the pattern reflects genuine belief or strategic positioning.

We don’t need AI regulation — leave safety to us, Nvidia’s Jensen Huang says — TechCrunch — Huang argued AI is hardware and software like any other, so safety should be engineered by product makers rather than mandated by government.

AI and data centers are incredibly unpopular in every poll — The Verge — A new NYT/Siena poll of 1,503 likely voters found 61 percent oppose data center construction for AI, adding political fuel to the slowdown debate.

Early Anthropic hire and former METR COO found a way to rein in rogue AI agents — TechCrunch — The AI Underwriting Company (AIUC) built a third-party audit and certification layer that uses AI agents to stress-test an enterprise’s own agents and report where trust is and isn’t warranted.

AI labs want in-house auditors — but maybe they should shut the front door first — TechCrunch — A simpler structural fix for rogue-agent risk may be more effective than internal auditor programs.

🧠 Frontier Models

Google released Gemini 3.8 Live and 3.8 Live Extended Thinking — Simon Willison / Google Google shipped two new speech-to-speech models comparable to OpenAI’s GPT-Live family, reportedly priced at less than half the cost of competing live-dialogue APIs. Gemini 3.8 Live Extended Thinking adds parallel reasoning for complex voice tasks; both are usable immediately via browser with a voice preset and optional system prompt. A companion Gemini 3.5 Transcribe model was released for transcription-specific workloads.

Introducing System One Models & Jev — TypeSafe AI TypeSafe AI introduced a new model class purpose-built for fast, structured software decisions rather than open-ended conversation. The flagship, Jev, matches large frontier models on “System One” tasks while running two orders of magnitude faster and being architecturally unable to hallucinate structured outputs. Available now in early access.

Introducing Odyssey-3: A General-Purpose Physical Intelligence — Odyssey Systems Odyssey-3 is a foundation world model — an autoregressive diffusion transformer trained to simulate highly diverse real-world scenarios. It learns physics, dynamics, cause-and-effect, and human behavior, enabling a single model to control robots, humanoids, vehicles, and drones or train other AIs in simulation.

Nature Is Our Learning Environment — Periodic Neon — Periodic — A scientific-domain specialist trained via midtraining and reinforcement learning on lab data that surpasses GPT-6 Astra and Claude Fable 5.1 on FrontierXRD at lower cost per analysis; now deployed in labs working on superconductors and magnets.

Shanghai AI Lab releases free 744B agent beating GPT-5.6 Sol on BrowseComp — Build Fast With AI — A massive open-weight model outperforming OpenAI’s current reasoning leader on web-based agentic benchmarks is now available on Hugging Face.

🛠️ Developer Tools & Platforms

Claude comes for Gemini with its own take on Docs and Slides — The Verge Anthropic launched Docs and Slides directly inside Claude, allowing document and presentation creation from chat with export and sharing. Simultaneously, Anthropic unified its separate Claude chat and Cowork products into a single “one Claude” interface — simplifying the product surface while preserving collaboration features. Currently rolling out to Pro and Max plan subscribers.

Google will now let any AI agent run your smart home — The Verge Google launched Google Home MCP in early access, opening its smart home ecosystem to Claude, ChatGPT, and other third-party AI agents via the standardized Model Context Protocol. Agents can now control devices, review camera summaries, and access activity logs using natural language.

Have it both ways: stay discoverable in search while disallowing AI training — Cloudflare Blog — Cloudflare’s new “Disallow AI Training” setting lets website owners block training crawlers while remaining indexed for search; Apple, Google, and Microsoft have committed to honoring the signal.

I made my website charge AI agents a penny per page — then watched Claude pay it — suganthan.com — An experiment with the x402 pay-per-crawl protocol demonstrates that Claude will actually execute a micropayment to access content, offering an early look at agent-native content monetization.

OpenResearch (GitHub Repo) — alphaXiv — A local-first workspace that turns Claude Code or Cursor into a research agent capable of autonomously proposing ideas, running experiments, and analyzing results over SSH, Kubernetes, or Hugging Face Jobs.

Open Code Review (GitHub Repo) — Alibaba — Alibaba’s internal AI code review assistant (now open-source), combining deterministic engineering with an LLM agent for line-level review precision across millions of defects found in production.

Claude Code 2.1.273 changelog — Anthropic — Adds gateway hint headers (x-claude-code-request-class, agent type, tool durations, compaction state) for LLM gateway operators, plus MCP server disconnect notifications and automatic reconnection.

💰 Business & Funding

OpenAI Considers Pre-IPO Funding Round at More Than $1.2 Trillion Valuation — The Wall Street Journal OpenAI is in early discussions with investors about a financing round ahead of its anticipated IPO next year. The company has crossed one billion active users and 200 million business customers. In a separate interview, Noam Brown said recursive self-improvement is OpenAI’s clearest near-term priority and that AI may surpass his own research intuition within one or two model releases.

Introducing Meta One — Meta — Meta launched a subscription service starting at $2.99/month that bundles enhanced AI and self-expression features across Instagram, Facebook, and WhatsApp, with individual, creator, and business tiers.

Former Infosys chief’s AI startup nabs another $53M — TechCrunch — The Palo Alto startup has signed multiple seven-figure enterprise contracts within months of its initial launch.

The AI graveyard: a running list of projects and startups that didn’t make it — TechCrunch — Relay, an AI workflow tool, is the latest addition, shut down by competition from larger platforms absorbing its feature set.

🦾 Robotics & Physical AI

Agility’s new humanoid robot will stop, squat to avoid harming human coworkers — Ars Technica Agility Robotics debuted a humanoid specifically engineered for workplaces with human coworkers: it detects nearby people and autonomously steps aside, stands still, or squats to a seated position to avoid contact. Early access arrives in the first half of 2027, with general availability to follow.

OpenArm (GitHub Repo) — Enactic — An open-source humanoid arm for teleoperation, imitation learning, and physical AI research.

Generated by claude-sonnet-4-6 · 2026-09-16T10:00:00Z