The Tech Download Can We Control AI? DeepMind’s Plan for Responsible AI
Jan 29, 2026
Tom Liu, VP for Frontier AI Global Affairs at DeepMind, works on international AI policy and harmonized standards. Dawn Blockswitch, Senior Director for Responsible Development at DeepMind, leads safety frameworks and ethical deployment. They discuss building safeguards, stress‑testing and red teaming, global regulatory coordination, transparency and independent evaluation, and the balance between rapid innovation and safety.
AI Snips
Chapters
Transcript
Episode notes
Design Responsibility Into Development
- DeepMind embeds responsible governance, research, and impact into development from the start.
- They treat near-term harms and long-tail frontier risks as interconnected problems to manage together.
Pair Red Teaming With Structured Tests
- Use both unstructured red teaming and structured evaluations to find model failures.
- Turn red-team discoveries into repeatable tests and fix models before wider release.
Safety And Speed Are Complementary
- Safety and speed must be integrated rather than traded off against each other.
- Industry forums like the Frontier Model Forum help align labs on risks and best practices amid competition.