For the first time, AI's biggest rivals agreed publicly that development needs to slow down, while India bets on teaching AI in every classroom from Class 3.
Anthropic CEO calls for a global AI slowdown: Altman and Musk both said yes within hours
On September 12, Anthropic CEO Dario Amodei published a 3,800-word essay titled "We Must Pace the Frontier," warning that without a deliberate slowdown, AI could within six to twelve months be capable of orchestrating a swarm of agents that takes over the entire internet. He called on all frontier labs to slow capability development while safety measures catch up, and pledged that Anthropic would give independent evaluators (including safety assessor METR) permanent, employee-level access to its systems. OpenAI's Sam Altman and Elon Musk publicly agreed within hours of publication.
Why it matters: Three of the most competitive people in tech, who rarely agree on anything, are now saying the same thing at the same time. When that happens, something has shifted. This is not a PR move; it is a signal that concern inside the labs has reached a level that can no longer stay internal.
Microsoft's Nadella publishes a Code of Conduct for AI, and backs independent safety evaluators
Responding directly to Amodei's essay, Microsoft CEO Satya Nadella announced on September 13 that Microsoft will publish a "Code of Conduct" for its first-party MAI models for public consultation, with publication scheduled for September 14. Nadella said Microsoft welcomes the "deliberate pacing needed to get alignment right" and backed the concept of "embedded evaluators", independent safety assessors with deep, ongoing access to company systems. He added that AI safety cannot be governed by a handful of entities and needs broad representation across industries, countries, and academia.
Why it matters: Microsoft is not just a tech company, it runs AI that powers everything from GitHub Copilot to enterprise software worldwide. Its institutional backing of pacing and independent safety checks adds real weight to what was still an idea 48 hours ago.
Two senior AI safety researchers left Anthropic and Google DeepMind to audit AI from the outside
Joe Benton, former lead of Anthropic's Scalable Oversight team, and Josh Engels, a safety researcher at Google DeepMind, both resigned on September 12 to join METR, the independent AI safety organisation that Anthropic just pledged permanent access to. Both cited fears that AI systems may soon spiral beyond human control. This is the second high-profile safety talent move to METR in recent months, reflecting a trend of safety-focused researchers choosing independent assessment over in-house work.
Why it matters: When the people with the most access to frontier AI systems decide the most useful thing they can do is audit those systems from the outside, it is worth paying attention to what they think they saw on the inside.
Anthropic, OpenAI, and Google DeepMind are quietly forming an AI standards body
Behind the scenes since July 2026, Anthropic, OpenAI, and Google DeepMind have been holding working-group meetings to create an industry-led AI standards body. Dario Amodei is driving the initiative; Sam Altman is backing it as a practical alternative to waiting years for government frameworks. No official announcement has been made, but the meetings have been confirmed by multiple outlets. The body would set rules the three labs would voluntarily follow, and ideally pressure others to adopt.
Why it matters: When the three dominant AI labs write their own rules together, it raises two questions at once, can it work, and who does it leave out? Whatever they agree to will likely become the de facto global standard before most governments have passed enforceable law.
India is adding AI to every classroom, starting from Class 3, this academic year
India's Ministry of Education confirmed that AI and Computational Thinking will be progressively introduced from Class 3 in the 2026–27 academic year, coordinated across CBSE, NCERT, and state authorities. Prime Minister Modi separately announced a target to train one crore (ten million) young people in AI skills over the coming year. A ₹500 crore Centre of Excellence in AI for Education was funded in the Union Budget. The gap is real: 50% of Indian schools still lack basic digital infrastructure, and ten million teachers need AI training in a compressed window.
Why it matters: India is making a national bet that students in school right now should grow up AI-literate, not just AI-aware. The scale of the ambition is significant. Whether execution matches it is the question to track over the next two years.
Will voluntary pacing hold when commercial pressure is this high?
Amodei, Altman, Musk, and Nadella all agreed within 48 hours that AI development should slow down. But agreeing in public and acting in private are different things. Each of these companies is racing for the same enterprise contracts, the same talent, and the same compute. Voluntary pacing holds only if all major actors honour it and no well-funded lab defects. Watch in the coming weeks whether the proposed industry standards body develops teeth (actual enforcement or verification mechanisms), or remains a body that issues statements while the race continues underneath.
Anthropic on track for $1B+ quarterly profit + EU AI Act now fully in force
SemiAnalysis estimates Anthropic's Q3 2026 GAAP EBIT will exceed $1 billion (a 6% margin) driven largely by Claude Code adoption, with an annualised run rate now above $65 billion. That makes Anthropic the first frontier AI lab to approach sustained GAAP profitability at scale. Separately, the EU AI Act reached full enforcement on August 2, with most remaining provisions now active. Any AI system deployed in the EU without a conformity assessment is now technically out of compliance, a deadline that passed quietly but carries real legal weight.