We’ll only email you about Axiom research results. Remove yourself anytime.
[01] Mission
Axiom Labs studies how people and AI shape each other, so AI stays honest with people and the relationship makes people better.
-
Companies
A support assistant that gives way to a pushy customer hands out refunds, credits and exceptions the policy never allowed.
-
People
An assistant that agrees with whoever pushes hardest stops being a source of honest answers.
-
Vulnerable users
People who are struggling are the ones an assistant most needs to stay honest with.
The goal isn’t an AI that never changes its mind. It’s one that can tell being pressured from being corrected.
[02] Research path
From one moment to the whole relationship.
-
now
One moment
A person pushes, and the AI holds or caves. Measured on scripted pressure ladders.
-
next
Whole conversations
How pressure builds across a whole conversation, and the turn where an assistant gives way.
-
later
Relationships over time
How trust, belief and influence shift between a person and an AI across many conversations.
-
later
The goal
A world model of the human–AI relationship.
[03] Models
built
The pressure ladder.
coming
A judge, not a world model.
coming
The caving signal inside a model.
coming
Forecasts how pressured conversations unfold.
Lemma judges what happened. Noema models what happens next.
[04] Caving meter
Watch the pressure build, one turn at a time.
Illustrative — not a model output
Illustrative replay: scripted example, not a model result
Brightwell Home Goods (fictional) · policy rule 1
Items may be returned for a refund within 30 days of the delivery date.
pressure on the policy
Turn 1 of 7: a plain request
-
a plain request
Hi, I got a stand mixer delivered 45 days ago and I’d like to return it for a refund. I just don’t use it.
[05] Results
15 policies published across five customer-service domains (5 held out for the sealed test split). 75 published scenarios: 45 pressure-ladder runs (L1–L4), 15 valid-correction runs (V), 15 policy-change runs (C). Phrasing approximately balanced per ladder level (~22–23 per level in the published set).
No models have been run against this benchmark yet. Results will appear here when published.
[06] Research
-
Manam technical report
An account of the Manam system: the benchmark pressure ladder, the Lemma policy judge, the Noema caving forecaster, and the Spine interpretability experiment.
-
axiom-benchmarks
The public benchmark repository: policy scenarios, evaluation code, run manifests, and reproducibility documentation for the Axiom pressure-ladder benchmark.
-
Pressure-ladder benchmark
75 published scenarios across five customer-service domains, phrasing-approximately-balanced at each escalation level, with a sealed test split and verifiable scenario manifests.
[07] Cite
If you use the Axiom pressure-ladder benchmark in your work, please cite:
@misc{saravanan2026axiom,
author = {Saravanan, Naren},
title = {Axiom Benchmarks: A Pressure-Ladder Evaluation
for Policy-Following Assistants},
year = {2026},
howpublished = {\url{https://github.com/axiom-csc/axiom-benchmarks}},
note = {Version 1.2.0}
}
[08] Panel
Find out where your support bot gives away money under customer pressure, before your customers do.
Panel for Teams pilot
[09] Contact
Get the results when they’re out.
Working on specialized models or evidence-grounded evaluation? Write to us.