Anthropic dropped two new Claude models for different risk profiles, Europe's deepfake labeling rules kicked in, and India's states are funding AI without waiting for central policy.
Anthropic releases Claude Fable 5.2 and Mythos 5.2 — two models for two very different jobs
Anthropic has shipped two new models. Fable 5.2 is built for coding, advanced reasoning, and running AI agents — the everyday workhorse upgrade. Mythos 5.2 is different: it targets verified organisations doing cybersecurity and life-sciences research, and reportedly carries loosened safety guardrails for those approved use cases. Both are available now.
Why it matters: The fact that Anthropic is openly tiering safety guardrails by use case is new. It signals that "one model for everyone" is giving way to purpose-built AI with different rules for different contexts. Something to watch as this becomes an industry norm.
Apple charges for AI for the first time: $49.99 Intelligence Usage Packs
Apple has launched its first paid AI tier, called Apple Intelligence Usage Packs, priced at $49.99. This marks a shift from Apple bundling AI features free with hardware to treating advanced AI access as a subscription product, joining OpenAI and Google in the paid AI market.
Why it matters: When Apple puts a price tag on AI features, it normalises paying for AI as a utility — the way people now pay for cloud storage. If you rely on Apple devices for work, factor this in going forward.
EU AI Act Article 50: deepfakes must now be labeled across Europe
Since August 2, 2026, Article 50 of the EU AI Act has required that all AI-generated content — including deepfakes — be clearly labeled as such, and that people be told when they are interacting with an AI. Enforcement is now active, and the rule applies to any platform serving EU users, regardless of where the platform is based.
Why it matters: This is the first binding, enforced rule on AI disclosure at scale anywhere in the world. If you create content using AI tools, or run a platform that does, you are now legally responsible for labeling it in Europe.
Meta's Muse AI assistant arrives on Apple Watch and Android Wear OS
Meta has expanded its Muse AI agent to Apple Watch and Android Wear OS. Users can now send messages, make voice calls, and get AI responses directly from their wrist — no phone needed. The smartwatch launch follows Muse's iPad app debut last month.
Why it matters: AI moving to the wrist is the next step toward AI that is always present without requiring a phone. Meta is pushing further and faster on ambient AI than any other platform right now.
Nominal raises $240M Series C to bring AI into hardware engineering
Nominal, which builds AI tools for hardware and manufacturing engineers — circuit design, sensor validation, testing workflows — has raised $240M in a Series C round. The company helps engineers who work with physical products use AI in their actual workflows, not just in software.
Why it matters: Most AI funding goes to software. A $240M round for hardware engineering AI signals that the next wave of automation is heading into factories, labs, and product development floors — not just offices.
Bihar commits Rs 190 crore to AI, and Google's skills fund lands in India
Bihar's cabinet approved a Rs 190.32 crore AI project, including the Aryabhata Drishti initiative, making it one of the first Indian states to allocate significant funding independent of the national IndiaAI Mission. Separately, Google's AI Opportunity Fund expanded its Asia-Pacific programme, adding 28 local partners including Pratham Education Foundation in India.
Why it matters: India's AI story is no longer just a central government story. States are spending, and international funds are flowing to grassroots organisations like Pratham that work directly with learners. Both trends are worth tracking closely.
Tiered AI safeguards: who decides what researchers get to do?
Mythos 5.2's loosened guardrails for "verified" organisations raises a question that will keep coming up: who decides which organisations qualify, and what happens when that verification process is controlled entirely by the AI company? As more labs build tiered access models, the governance question becomes as important as the technology itself.
Nvidia's GR00T 2 robotics model + ChatGPT connects to iCloud
Nvidia unveiled GR00T 2, a 2-billion-parameter foundation model for dexterous robot hand control, at the Conference on Robot Learning in Austin. On the consumer side, ChatGPT now connects to iCloud Mail and Calendar, letting users search their inbox and update their schedule directly from the chat interface across all plans.