Redwood Research blog

Redwood Research blog

Home
Podcast
Chat
Reading List
Archive
About
Blocking live failures with synchronous monitors
A common element in many AI control schemes is monitoring – using some model to review actions taken by an untrusted model in order to catch dangerous…
 • James Lucassen and Adam Kaufman
Reward-seekers will probably behave according to causal decision theory
They'd renege on non-binding commitments, defect against copies of themselves in prisoner's dilemmas, etc.
 • Alex Mallen
AI's capability improvements haven't come from it getting less affordable
AI inference is still cheap relative to human labor
 • Anders Cairns Woodruff
Are AIs more likely to pursue on-episode or beyond-episode reward?
RL would encourage on-episode reward seeking, but beyond-episode reward seekers may learn to goal-guard.
 • Anders Cairns Woodruff and Alex Mallen
The case for satiating cheaply-satisfied AI preferences
Some unintended preferences are cheap to satisfy, and failing to satisfy them needlessly turns a cooperative situation into an adversarial one.
 • Alex Mallen

February 2026

Frontier AI companies probably can't leave the US
The executive branch can, and probably would, block a frontier AI company's departure.
 • Anders Cairns Woodruff
Announcing ControlConf 2026
Berkeley, April 18-19
 • Buck Shlegeris
Will reward-seekers respond to distant incentives?
Reward-seekers are supposed to be safer because they respond to incentives under developer control. But what if they also respond to incentives that…
 • Alex Mallen
How do we (more) safely defer to AIs?
How can we make AIs aligned and well-elicited on extremely hard to check open ended tasks?
 • Ryan Greenblatt and Julian Stastny
Distinguish between inference scaling and "larger tasks use more compute"
Most recent progress probably isn't from unsustainable inference scaling
 • Ryan Greenblatt

January 2026

© 2026 Buck Shlegeris · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture