How Google DeepMind Tests AI Before It Goes Wrong

Two of Google DeepMind’s leaders—Dawn Bloxwich (Responsible Development & Innovation) and Tom Lue (VP, Frontier AI Global Affairs)—open the playbook on frontier safety and policy. Bloxwich explains how DeepMind blends structured evaluations with red‑teaming by experts and jailbreakers, and how its Frontier Safety Framework addresses severe risks (from CBRN and cyber to loss of control and socio‑affective concerns). Lue maps the regulatory landscape across the EU, US and Asia, arguing for harmonized standards that safeguard without stifling innovation. Together, they show what responsible development looks like before and after deployment—model cards, third‑party testing and the industry forums trying to align best practices.

CHAPTERS:

DeepMind's Holistic Strategy for Safe AI Development
11

Addressing Agentic AI, Ethics and Independent Evaluation
20

Reflecting on AI Stress-Testing and Balancing Speed with Safety
22

Navigating International Standards and Geopolitical Competition in AI
29

Engaging Governments, EU Laws and AI Content Business Models
35

Predicting Agentic AI, Labor Impacts and Scientific Breakthroughs
39

The Paradox of Regulation and US Federal AI Policy

#CNBC #CNBCPodcast #TheTechDownload

-----

Subscribe: @cnbci

CNBC International Live: @cnbcinternationalive