AI Pulse: New models, real-world pilots, and the next governance test

EMXETF has announced the China AI Tigers LLM ETF, listed as TGRZ on Nasdaq, offering investors a way to target companies associated with leading Chinese AI models. For builders, it is another sign that model ecosystems are becoming investable themes—not just technical projects.

A physics AI startup founded by Anandkumar and Jenik reportedly declined a Bezos-backed offer and launched a neural operator model designed to handle 5 trillion data points in a single prompt. The practical question for technical teams is whether architectures built around scientific structure can make large-scale simulation more usable and efficient.

Amazon’s Mechanical Turk, launched in 2005 to outsource tasks that computers struggled with, is shutting down according to CNBC. Its trajectory is a useful reminder that “AI” products often depend on hidden human labor—and that changing economics can reshape those systems.

The Stowers Institute has expanded its AI initiative and appointed Charles McAnany as a second AI Fellow. His work uses machine learning to identify patterns in DNA and study gene regulation, showing how specialized domain knowledge remains central to meaningful scientific AI.

Kiley Prep Middle School in Springfield is beginning an AI classroom pilot while educators voice concerns about its use. For schools, the hard part is not simply access to tools; it is setting expectations around teacher judgment, student privacy, learning goals, and evidence of benefit.

Insilico Medicine’s interim first-half 2026 results point to a commercialization roadmap backed by revenue in the three-digit-million-dollar range, according to the announcement. AI drug-discovery teams will be watching how this business progress connects computational methods with clinical and commercial outcomes.

Google has expanded Gemini Enterprise with tools aimed at lawyers and law firms, intensifying competition for professional-services work. Legal teams evaluating these systems should focus not only on drafting speed, but also on review processes, confidentiality, traceability, and who remains accountable for the final advice.

Alibaba has launched Wan 3.0, extending its generative AI efforts into video. For creators and product teams, another major model entrant could mean more choice—but also a growing need to compare controllability, consistency, rights management, and cost rather than judging tools on demos alone.

An Alabama investigation is described as opening a new regulatory front: examining how companies contain powerful AI models, not only what those models say to users. That shift matters for builders, because safety expectations may increasingly include system design, access controls, monitoring, and deployment practices.

Revolut has established Revolut Research inside its broader AI department to work on machine-learning systems for financial services. A dedicated research group can help a fintech build capabilities around its own problems, but production value will depend on reliability, governance, and measurable improvements for customers and staff.