PAIR-SAFE: A Paired-Agent Approach for Runtime Auditing and Refining AI-Mediated Mental Health Support

Dong Whi Yoo; Eshwar Chandrasekharan; Jiwon Kim; Koustuv Saha; Violeta J. Rodriguez

RECORD

PAIR-SAFE: A Paired-Agent Approach for Runtime Auditing and Refining AI-Mediated Mental Health Support

Jiwon Kim, Violeta J. Rodriguez, Dong Whi Yoo, Eshwar Chandrasekharan, Koustuv Saha

Human-Computer Interaction · Submitted 19 Jan 2026

abstract compiled · 2026-10-26 · no language-model review

The authors' main claim Our study reveals that paired-agent auditing can provide clinically grounded, transparent oversight for AI-assisted mental health support.

how it is stated · asserted 0.58 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID2601.12754 (version 2)

Pith ID7G6AJN2W, Pith's permanent record of this paper and its review

CitationDong Whi Yoo, Eshwar Chandrasekharan, Jiwon Kim, Koustuv Saha, Violeta J. Rodriguez. "PAIR-SAFE: A Paired-Agent Approach for Runtime Auditing and Refining AI-Mediated Mental Health Support." arXiv:2601.12754 (2026). https://pith.science/paper/7G6AJN2W

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-10-26: 1 request, 6,117 input tokens.

What the abstract itself states, as probabilities: claims first 0.06 · states limitation 0.18 · human participants 0.17 · controlled comparison 0.33 · randomized assignment 0.02 · states quantitative result 0.11 · compares with prior results 0.50 · mentions code or data availability 0.03.

The abstract, sentence by sentence

arXiv listing abstract v2. Results and the main claim are underlined; show every role.

LLMs are increasingly used for mental health support, yet they can produce responses that are overly directive, inconsistent, or clinically misaligned. Most risk mitigation approaches rely on implicit alignment through training or prompting, offering limited runtime accountability. We introduce PAIR-SAFE, a paired-agent framework that integrates a Responder agent with a Judge agent grounded in the clinically validated Motivational Interviewing Treatment Integrity (MITI-4). The Judge audits each response and issues \allow{} or \revise{} decisions to guide real-time refinement. We evaluate interactions using a support-seeker simulator derived from human-annotated data. The Judge-supervised interactions show significant improvements in key MITI dimensions, including Partnership, Seek Collaboration, and overall Relational quality. Our quantitative findings are corroborated by expert evaluation, highlighting nuances of runtime supervision. Our study reveals that paired-agent auditing can provide clinically grounded, transparent oversight for AI-assisted mental health support.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	LLMs are increasingly used for mental health support, yet they can produce responses that are overly directive, inconsistent, or clinically misaligned.	background 0.99	0.04	0.02	asserted 0.82
s1	Most risk mitigation approaches rely on implicit alignment through training or prompting, offering limited runtime accountability.	gap 0.89	0.04	0.02	asserted 1.00
s2	We introduce PAIR-SAFE, a paired-agent framework that integrates a Responder agent with a Judge agent grounded in the clinically validated Motivational Interviewing Treatment Integrity (MITI-4).	contribution 0.86	0.86	0.06	asserted 1.00
s3	The Judge audits each response and issues \allow{} or \revise{} decisions to guide real-time refinement.	method 0.98	0.76	0.04	asserted 0.99
s4	We evaluate interactions using a support-seeker simulator derived from human-annotated data.	method 1.00	0.95	0.02	asserted 0.99
s5	The Judge-supervised interactions show significant improvements in key MITI dimensions, including Partnership, Seek Collaboration, and overall Relational quality.	result 1.00	0.96	0.20	asserted 1.00
s6	Our quantitative findings are corroborated by expert evaluation, highlighting nuances of runtime supervision.	result 0.93	0.90	0.09	asserted 0.99
s7	Our study reveals that paired-agent auditing can provide clinically grounded, transparent oversight for AI-assisted mental health support.	interpretation 0.99	0.94	0.92	asserted 0.58

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.75.

What was examined, and what was not

sources read

arXiv listing abstract v2

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

LLMs are increasingly used for mental health support, yet they can produce responses that are overly directive, inconsistent, or clinically misaligned. Most risk mitigation approaches rely on implicit alignment through training or prompting, offering limited runtime accountability. We introduce PAIR-SAFE, a paired-agent framework that integrates a Responder agent with a Judge agent grounded in the clinically validated Motivational Interviewing Treatment Integrity (MITI-4). The Judge audits each response and issues \allow{} or \revise{} decisions to guide real-time refinement. We evaluate interactions using a support-seeker simulator derived from human-annotated data. The Judge-supervised interactions show significant improvements in key MITI dimensions, including Partnership, Seek Collaboration, and overall Relational quality. Our quantitative findings are corroborated by expert evaluation, highlighting nuances of runtime supervision. Our study reveals that paired-agent auditing can provide clinically grounded, transparent oversight for AI-assisted mental health support.

Human review

How Pith checks papers → · Full audit of this paper →

References

How the literature uses it

5 arXiv papers cite this in 6 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 1 sentence without a confident label

Uses its method or tool 1 paper

Following prior work on LLM-as-judge evaluation [ref] , we evaluate generated explanations with a structured rubric prompt.

Causal Stories from Sensor Traces: Auditing Epistemic Overreach in LLM-Generated Personal Sensing Explanations · LLM-Based Automated Evaluations · 0.98

Cites it as background 4 papers

Existing approaches have primarily relied on prompt engineering, supervised fine-tuning, and reinforcement learning-based alignment methods [ref] .

LLUMI: Improving LLM Writing Assistance for Mental Health Support with Online Community Feedback · Related Work · 0.83
LLMs inherit biases and toxic patterns from their training data, leading to concerns about fairness, safety, and harmful content generation [ref] .

Prompt PerturbAItions: Mechanisms of LLM Response Inconsistencies · Related Work · 0.92
Multi-agent LLM systems (MAS) are increasingly deployed in coordination-heavy domains such as scheduling, care coordination, financial workflows, and enterprise pipelines [ref] , where they communicate through compressed handoffs: one agent's summary, forwarded message, or memory entry becomes the next agent's ground truth.

Facts Without Rules: Boundary Metadata Collapse in Multi-Agent LLM Handoffs · Introduction · 0.82

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.