Pretraining Safety w/ Ethan Roland
Download MP3What if the safest AI models weren't built by adding guardrails after training, but by shaping what gets learned in the first place? Ethan Roland, senior alignment researcher at AE Studio and first author on an ICML 2026 spotlight paper, joins Jacob to talk about gradient routing, a technique that routes dangerous capabilities into isolated parts of a model's architecture where they can be locked or removed entirely. They get into the absorption effect, KYC-style access control frameworks, and what it would actually take for frontier labs to adopt this kind of work before it's needed rather than after.
Chapters
Links
Below are the most important links for this episode. For more, visit the episode page on Kairos.fm.
Chapters
- (00:00) - Introduction
- (06:39) - Inside AE Studio
- (15:26) - China & the Alignment vs. Controllability Framing
- (18:23) - Data Filtering & Gradient Routing (Aside)
- (30:39) - Mixture of Experts Explained (Aside)
- (36:25) - Why Pre-Training Interventions Are Rare
- (42:43) - Ethan's Theory of Change
- (56:17) - Access Control Governance and KYC (Aside)
- (01:04:47) - The Researcher's Role in Policy Advocacy
- (01:11:38) - Speed Round
- (01:27:55) - Outro
Links
Below are the most important links for this episode. For more, visit the episode page on Kairos.fm.
- Ethan's website
- The paper landing page
- Preprint - Gradient Routing: Masking Gradients to Localize Computation in Neural Networks
- ICLR paper and webpage - Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
- Wikipedia article - CBRN defense
- NTI tutorials on bioweapons and nuclear testing
Creators and Guests
