The BaseResearchAI Alignment Forum
A Simpler and Better Debate Protocol for Scalable Oversight
Zhiyang Xun, Jiawei Li and Jonah Brown-Cohen present FBS Debate, a theoretical protocol for AI debate, an approach to supervising systems whose work is too complex for people to judge directly. They claim stronger correctness guarantees than the current best protocol and prove that no protocol treating human judgment as a black box does better.
Why it matters
The same thing, for different readers.
What it means for you
This is early theory about how one AI might check another's work when people cannot, not a tool in use today.
For researchers
Worst-case rather than average-case guarantees for debate, with a lower bound that depends on treating human judgment as a black box, so progress may need new assumptions about judges.
The question it raises
Whether an argument relying on fuzzy human judgment at each step can be trusted if small errors flip the outcome.
In their own words
Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.
From the source’s own text.
“Furthermore, we show that FBS Debate is optimal on every individual question: unless further assumptions are made about human judgments, no other protocol can find a flawed sub-claim asymptotically more efficiently.”
“In FBS Debate, honest play is the best strategy for each debater, no matter what the other does: if the original claim is correct, an honest Alice always wins, and if it is incorrect, an honest Bob wins with overwhelming probability.”
“Our lower bounds only apply when human judgment is modeled as a black box.”
How the finding was reached
In our words, from the source. Weigh the finding by how it was made.
- What was done
- A theoretical protocol and proofs for recursive AI debate on problems that decompose into sufficiently stable sub-claims, with human judgment modeled as a black box.
- What it found
- FBS Debate gives worst-case correctness on each question, for a stronger equilibrium concept than prover-estimator debate, and no protocol treating human judgment as a black box can find a flawed sub-claim asymptotically more efficiently.
- What it does not show
- Applies only to problems with sufficiently stable decompositions and to protocols that treat human judgment as a black box; Bob's repeated rollouts cost inference-time computation; no experiments with real models are described in the post.
- Status
- A preprint: not yet peer reviewed
The record
- Questions
- Can AI systems act beyond what people intended or allowed? · Are AI companies putting stronger safeguards into practice?
- People
- Zhiyang Xun (Wrote it) · Jiawei Li (Wrote it) · Jonah Brown-Cohen (Wrote it)
- What we read
- We read the whole source.
- Found by
- The Radar, run of October 17, 2026, via
ai-alignment-forum(tier 1: the source’s own channel or an official record) - Cite this
- AI Alignment Forum, “A Simpler and Better Debate Protocol for Scalable Oversight”, October 16, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/alignment-forum-2026-10-16-simpler-better-debate-protocol