The three biggest AI labs are negotiating a self-regulatory standards body, a 600B-parameter model opened its API, and India's AI liability law heads to parliament.
OpenAI, Anthropic & Google are building an AI standards body, modelled after Wall Street's regulator
The three leading AI labs have been meeting since July to design a self-regulatory organisation modelled after FINRA, the body that oversees US brokers. OpenAI CEO Sam Altman has publicly endorsed the idea, saying a testing and auditing body should exist, but must be built by labs themselves, not government. Formal talks became public in mid-September and are accelerating.
Why it matters: This is the industry's biggest attempt yet to write its own rules before lawmakers do. Whether it becomes meaningful accountability or a clever way to delay harder regulation is the question worth tracking.
StepFun's Step 5 Preview API opens: a 600-billion-parameter model anyone can now call
Chinese AI lab StepFun opened API access to Step 5 Preview on September 20. It is a 600B-parameter sparse mixture-of-experts model with only 27B parameters active per token and a one-million-token context window. Early benchmarks show it outperforming several frontier alternatives on reasoning and long-document tasks.
Why it matters: A model this capable with open API access raises the ceiling for anyone building AI products, and raises fresh questions about how a self-regulatory body would handle models of this scale released outside the US.
OpenAI's Agents API enters public beta, the infrastructure powering ChatGPT agents is now yours
OpenAI has opened its Agents API to all developers in public beta. It exposes the same managed harness used internally for Codex-style agents: four core building blocks (agent, environment, session, and events), that let developers create, run, and monitor agents at scale without building the orchestration layer themselves.
Why it matters: Previously, production-grade AI agents required significant custom infrastructure. This lowers the barrier significantly for startups and solo developers who want to ship agent-based products.
India's Digital India Act: strict liability for AI-generated harm heads to parliament
India's parliament is set to review a strict liability clause in the Digital India Act that would hold generative AI developers legally accountable for harms caused by their systems. IT Minister Ashwini Vaishnaw, speaking at Semicon India, signalled that regulation has a role and urgency is rising. The Ministry of Electronics is examining the formation of an AI Governance Group.
Why it matters: India moving from voluntary measures to binding liability law is a significant shift. If passed, it could become the most stringent AI liability framework in Asia, and one that directly affects international AI companies operating in India.
Two US state AI safety laws take effect January 1, 2027, companies have 100 days to prepare
New York's Responsible AI Safety and Education (RAISE) Act and Illinois's AI Safety Measures Act (IAISMA) both become enforceable on January 1, 2027. Both target "frontier models": systems trained using more than 10 to the power of 26 computational operations. Separately, the CART Act's transparency rules for synthetic digital content kick in on October 1, 2026.
Why it matters: The US has no federal AI law, but states are moving independently. Any company using or building powerful AI systems now has legal deadlines to plan around, not just voluntary guidelines.
Assort Health hits $1.2B valuation with $120M raise: AI-native healthcare crosses another milestone
Assort Health, which builds AI-powered patient intake and clinical workflow tools, closed a $120M Series C at a $1.2B valuation, bringing total funding to over $220M. The raise reflects continued strong investor appetite for AI applied to healthcare operations, a category that saw over $300B in global AI investment in the first half of 2026.
Why it matters: Healthcare AI is no longer a future bet, it is producing billion-dollar companies. If you work in medicine, nursing, hospital administration, or health policy, the AI tools reshaping your sector are already fully funded.
Who writes the rules for AI, the labs or the lawmakers?
The self-regulatory standards body talks between OpenAI, Anthropic, and Google are the most important governance story of the month. The precedent matters enormously: if the labs set the standards, they control the audit criteria, the penalties, and the pace of change. If governments step in instead (as India, the EU, and US states are all signalling) the power shifts. The next few weeks of negotiation will shape which of those futures becomes real.
India's data centre tax breaks vs. AI regulation tension + EU AI Act fully in force
India's 20-year tax holidays for data centres are drawing scrutiny as the country simultaneously pushes stricter AI regulation, experts argue fiscal incentives for compute may clash with proposals to tax data processing power. Meanwhile, the EU AI Act is now fully in force across all member states, meaning high-risk AI systems in healthcare, hiring, and education face active compliance obligations today, not next year.