AI Safety Wars, Royal Summits, and 100 GW of Grid Hunger
Top AI labs pump the brakes, King Charles gets nervous, and Big Tech wants to swallow 100 GW of power grid capacity.
Subscribe free →The AI Safety Slowdown Is Now an Official Movement — With Infighting
After a summer of rogue AI agents and dire researcher warnings, leading US labs are publicly calling to "pace the frontier" of bleeding-edge AI development. But Microsoft AI CEO Mustafa Suleyman just went on record saying the threats are real *and* that Anthropic is making things worse — even as his own company published a 37-page "Humanist AI Code of Conduct." The schism between labs on *how* to slow down is widening fast.
Why it matters: When the CEO of Microsoft AI is publicly attacking Anthropic's approach to safety, the AI industry's internal cold war has gone fully public.
King Charles Convenes a Private AI Summit — and Isn't Sold
King Charles hosted a closed-door summit with prominent AI leaders and U.K. government officials, emerging with notable hesitations about the technology's direction. Details from inside the room remain sparse, but the optics alone — a reigning monarch convening the AI industry — signal how far up the chain AI governance anxiety has traveled.
Why it matters: When heads of state start hosting private AI salons, the window for industry self-regulation is closing faster than anyone expected.
Google, Nvidia, and Anthropic Are Hunting for 100 Gigawatts of Grid Space
A new coalition — Google, Nvidia, Anthropic, and a company called Emerald AI — is working to unlock 100 GW of electrical grid capacity for new data centers. For context, 100 GW is roughly the entire current generating capacity of Germany. The group wants Emerald AI to identify viable grid locations that can absorb the construction of massive new AI infrastructure.
Why it matters: This makes the AI power crisis concrete: the industry isn't just asking for more electricity, it's trying to reshape national energy infrastructure to get it.
Huawei's Ascend 960DT Is Coming for Nvidia in Q1 2027
Huawei is accelerating the launch of its next-generation Ascend 960DT AI chip, targeting a Q1 2027 release as it pushes to close the gap with Nvidia in AI compute. The move is part of China's broader push to build a domestic AI chip ecosystem that doesn't depend on U.S.-controlled supply chains.
Why it matters: Every credible Nvidia competitor chips away at U.S. export controls as a containment strategy — and Huawei is now moving faster, not slower.
AI Agents That Make Phone Calls Are Now a Two-Horse Race
Instinct and Meta's Muse AI agent both added voice calling this week, letting users book restaurant reservations, cancel subscriptions, and handle other phone-based tasks without lifting a finger. The simultaneous launches suggest both companies are racing to claim "AI that does your annoying calls" before the feature becomes table stakes.
Why it matters: Phone-capable agents are the first AI feature that meaningfully replaces something people actually hate doing — and two serious players just planted their flags on the same day.
OpenAI and Anthropic's "Independent" Safety Evaluators Aren't That Independent
Both OpenAI and Anthropic have announced plans to embed external safety evaluators inside their labs — giving researchers unprecedented access to study frontier models. Researchers welcome the move but are flagging the obvious problem: evaluators funded by and housed inside the labs they're auditing face structural pressure to pull punches, and without regulation, there's no enforcement backstop.
Why it matters: Safety theater is worse than no safety process at all — it creates the illusion of oversight while the labs retain full control over what gets published.
Quick Hits
- →Pew Research surveyed 42,000 people across 37 countries: in 34 of them, majorities expect AI to destroy more jobs than it creates over the next 20 years. The Verge
- →Claude Code revamped its Projects feature to let users run multiple AI agents in parallel in the cloud, with a shared memory and a coordinator agent directing the whole operation. The Verge
- →Adecco Group is rolling out Salesforce's Agentforce Coworker to 27,000 employees across 40+ countries — one of the largest enterprise AI agent deployments announced to date. AI News
- →Waymo is back in San Antonio five months after one of its robotaxis got swept away in flooding — and separately, a different Waymo called the cops on two passengers it suspected of carrying a firearm. TechCrunch / The Verge
- →Base Labs (spun out of Baseten) launched an open-weight AI safety partnership with Hugging Face and Goodfire to publish training and monitoring methods for open models. TechCrunch
AI TLDR