Daily AI Brief

China's new Measures require platforms and providers to attach both visible and embedded identifiers (e.g., watermarks/metadata) to AI-generated text, images, audio and video. Enforcement begins now; expect downstream requirements for global brands operating in China.

A WSJ analysis argues Beijing's pragmatic, industry-focused AI approach (smaller DCs, more application spend) could deliver short-term gains while the West chases AGI.

Brands are preparing for chatbots/agents that browse, compare and even purchase on behalf of customers; optimisation shifts from SEO for humans to "AIO" for agents.

A fresh Verge column says agentic systems remain patchy for everyday users, with coding use-cases strongest so far.

Incumbents in legal research face a "new era of competition" as AI challengers scale.

Rules require explicit on-screen/audio notices and implicit metadata/watermarks; platforms must detect, label and moderate synthetic media.

Timing matters for data portability and contractual terms across cloud/AI services.

Anthropic has announced updated consumer terms with a 28 Sept deadline to accept and control training settings.

Researchers report ML approaches improving predictions for optical/semiconductor materials, promising better screening for devices.

WSJ reports VCs working weekends to secure AI deals; July saw a jump in US$100m+ rounds, with AI about half.

Reality-check on agent hype vs usable automation.

What brands should change now.

— Steve Hasker, CEO, Thomson Reuters
Source: Business Insider

Compliance workstreams should include watermark tech selection, vendor attestations and audit logs.

Contracts and data-sharing terms may need updates.