India doubled down on chips, California signed child chatbot safety law, China's 600B model API opened, and the EU's AI rulebook is now fully enforced.
China's 600B-parameter Step 5 just opened its API, weights go free October 15
Chinese AI lab StepFun announced Step 5 Preview on September 20 and opened API access the same day. The model has 600 billion total parameters, with 27 billion active per token, a one-million-token context window, and multimodal support for text and images. It targets long-horizon agent work (coding, financial analysis, and software engineering) at $1 per million input tokens. Model weights are scheduled to open for free download on October 15.
Why it matters: When those weights drop, any developer or researcher on the planet can download and run one of the world's most capable AI models without paying anyone. The gap between China and the US frontier is narrower than it was six months ago, and October 15 is the date to circle.
India doubles its chip budget to $13.5B: Applied Materials commits $5B
Prime Minister Modi inaugurated SEMICON India 2026 in New Delhi on September 17 and announced India Semiconductor Mission Phase 2, expanding the national chip-building budget from $8 billion to $13.5 billion. Applied Materials committed $5 billion over ten years. Lam Research pledged roughly ₹10,000 crore. India's domestically manufactured chips are already reaching their first customers.
Why it matters: AI needs chips. India's decision to build its own semiconductor supply chain is the most consequential long-term AI infrastructure investment the country has made. If successful, India could have the hardware independence to train its own frontier models within a decade.
India is testing AI that pays for you through UPI, automatically, no tap needed
India is piloting autonomous AI payments via UPI, the country's digital payments network with over 500 million active users. An AI agent would check your account, compare options, and complete a payment without you doing anything. This is agentic AI applied to daily financial life at a scale no other country has attempted.
Why it matters: This is the most practical, mainstream AI application India has announced to date. Half a billion people doing digital payments could soon have an AI handle routine transactions in the background. Watch how the Reserve Bank of India sets the guardrails.
California signs law making companies responsible for AI chatbot harm to children, before it happens
Governor Gavin Newsom signed a series of laws requiring tech companies to conduct risk assessments before launching AI chatbots that children could access. Companies found guilty of harming a child face fines of up to $1 million per child. This is the most specific state-level AI child safety law in the US, and it puts the legal burden on companies ahead of the harm, not after.
Why it matters: AI chatbots are already in study apps, tutoring platforms, and games children use every day. For the first time, a government has made companies legally accountable before a product goes live, not only after damage is reported.
EU AI Act is now fully in force. All AI companies selling in Europe must comply.
The European Union AI Act's final provisions took effect on August 2, 2026, completing a multi-year phase-in. All companies selling AI products in Europe must now comply with risk labelling, human oversight requirements, transparency rules, and safety standards. Fines for violations reach up to €35 million or 7% of global annual revenue, whichever is higher.
Why it matters: This is the world's most comprehensive binding AI law, and it applies to any company selling AI products in Europe, including Indian and US companies. The wait-and-see period is over; compliance is now mandatory and enforcement has teeth.
Deepfakes now account for 11% of all global fraud, hiring is the fastest-growing target
A new industry report puts deepfake attacks at 11% of all global fraudulent activity in 2026. Hiring is the hardest-hit sector: deepfake job candidate attempts jumped 1,300% between 2023 and 2024, and consumer-grade apps have now eliminated the technical barrier entirely. The EU AI Act now requires all AI-generated content to carry a machine-readable label.
Why it matters: If you hire people, receive voice calls, or join video meetings, the statistical likelihood that AI was used to deceive you has grown dramatically in the past year. Recognising what to look for is now a practical skill, not a paranoid one.
China's AI is closing the gap, and October 15 is the moment it goes fully open
Step 5 Preview's open weights are scheduled for October 15. Simultaneously, Naive AI (a Beijing startup founded in February 2026 by a Tsinghua professor) hit a $1.42 billion valuation after raising $400 million from Tencent and IDG in under a year. China is shipping frontier models faster, at lower cost, and with open weights. The US labs are still ahead on most benchmarks, but the margin is narrowing and the pace is accelerating. October 15 is the next date that matters.
Alibaba's Qwen team released Qwen3.8-Omni-Flash on September 18: a model that handles text, images, audio, and video together with a one-million-token context, available at no cost. On the education side, AI classroom tools are attracting real capital: MagicSchool AI has raised $59M, Brisk Teaching $15M+, and language-learning AI apps have crossed $650M cumulative, practical AI for classrooms is no longer underfunded.