Monday, September 28 September 28, 2026
An Anthropic researcher's public resignation letter invokes the Manhattan Project, Anthropic battles the White House over AI oversight, and Claude Opus 5.5 plus GPT-6 cut frontier model prices — all while China drafts a six-stage AI safety framework and J.P. Morgan calls the AI stock pullback a buy signal.
Good morning. It's Monday, September 28th, 2026. I'm your MorningAI briefing. Today the story isn't just about what AI can do — it's about what AI is doing to the people building it, regulating it, and paying for it. Let's get into it.
The week's biggest meta-story came from Anthropic. Researcher Jacob Coxon resigned this month in a public letter he titled, quote, "I Write This to Frighten You." Coxon drew parallels to the atomic scientists of the Manhattan Project — researchers who helped build something they later feared. The New York Times framed it as a moment of reckoning: when the people inside the labs are raising existential alarms, it's worth taking seriously. This isn't fringe concern anymore. It's coming from inside the house.
That safety tension is colliding with a regulatory battle. Anthropic is now publicly clashing with the U.S. administration over who should have oversight authority for AI systems. The company — arguably the most safety-focused lab in the game — is arguing for more federal guardrails, while the current administration prefers a lighter regulatory touch. This is a pivotal fight. The outcome shapes the entire compliance landscape for AI companies for the next decade.
On the model side, last week gave us a notable price war at the frontier. According to Smartkarma's weekly roundup, Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 both launched with meaningfully lower prices than their predecessors. Meanwhile, xAI's Grok 4.7 pushed deeper into coding channels, competing directly with Cursor and GitHub Copilot for developer mindshare. Cheaper frontier models means the barrier for enterprise adoption is dropping fast — and it's putting pressure on every layer of the AI stack below them.
Meanwhile, China is watching all of this and preparing its own response. The South China Morning Post reports that Chinese developers — who currently dominate the open-weight model ecosystem — are actively exploring how to make those models less dangerous. Beijing is reportedly considering a six-stage safety certification process. This matters because open-weight models like DeepSeek's family are used globally, and any safety framework China imposes will ripple through the international AI community.
The energy bill for all this compute is becoming impossible to ignore. A Geneva Solutions report out today highlights the strain AI data centers are placing on electrical grids worldwide. Power demand from AI infrastructure is outpacing the capacity utilities planned for, and that's triggering real friction with communities and regulators near major data center clusters. Sustainability is no longer a PR angle for hyperscalers — it's operational risk.
On the investment side, J.P. Morgan out this morning says the recent pullback in global AI stocks has actually created an attractive re-entry point. Valuations have reset, positioning is cleaner, and the bank sees room for a re-engagement — especially in semiconductor names. The U.S. Treasury Secretary also publicly linked AI to productivity gains this week, nudging the Fed to keep monetary policy flexible. Markets are watching AI as a macro signal now, not just a sector play.
One medical AI note worth flagging: a review published in News-Medical highlights a dangerous gap in AI health diagnostics. AI models are identifying promising biomarkers with high apparent accuracy — but many of those biomarkers are failing in large-scale clinical validation. The finding is a warning for anyone building or deploying medical AI: benchmark performance and real-world outcomes are still diverging in ways that matter for patient safety.
Here's today's business idea: an AI regulatory intelligence firm. With Anthropic fighting the administration, China drafting safety frameworks, and the EU already rolling out the AI Act, the compliance landscape is fracturing globally. A subscription service that translates regulatory developments into plain-language weekly action items for AI product teams — at a price point between a newsletter and a law firm — is a genuine gap in the market. Founders, compliance consultants, and enterprise AI teams all need this.
That's your Monday morning AI briefing. Safety signals are getting louder, the cost of frontier AI is dropping, and the regulatory map is being redrawn in real time. Stay sharp. We'll see you tomorrow.