- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
The paper studies privacy leakage in sequential LLM-agent pipelines. It defines global leakage through mutual information, derives a composition bound, and proposes MINE-based privacy regularization. Experiments on MedQA, FinQA, and PrivacyLens examine privacy, task utility, pipeline depth, and regularization strategies.
Critical strengths
The final-output attack evaluation directly addresses the stated threat model, using frozen agents and a disjoint attacker-training split. The reported reduction in sensitive-attribute recovery, with much of the benign task performance retained, provides useful evidence of a practical benefit from regularization.
Critical weaknesses
The central position-dependent interpretation is unsupported. Under the conditional data-processing and independence steps used in Appendix A.1, applying the chain rule directly to cumulative leakage yields a tighter additive bound without position-dependent weights. The exponential coefficients therefore reflect looseness in the derivation rather than an established positional effect. Proposition 1 restates these coefficients, leaving the claimed theoretical basis for prioritizing earlier agents unsubstantiated.
The baseline results are inconsistent across tables and metric definitions. Table 7 is labeled MedQA, but its LLaMA-3B baseline and proposed-method rows reproduce the corresponding FinQA results and disagree with the main MedQA table. The displayed PARI differences also do not reconcile with the stated metric formula and component values. These discrepancies leave the dataset provenance and comparative conclusions unresolved.
The protected information and attack task are insufficiently specified. Appendix A.3.4 lists private-prompt embeddings, routing states, and policy indicators without explaining their construction, sampling, or assignment to agents. The final-output evaluation introduces a four-class attack without defining the labels, class frequencies, attacker architecture, or split sizes. These omissions prevent reproduction of the central privacy evaluation and assessment of what sensitive information the reported protection covers.
The claimed privacy-utility advantage lacks a controlled comparison. The alternative interventions lack specified strengths and tuning protocols; the privacy-regularization baseline also lacks an explicit objective. Table 7 reports isolated operating points, while the trade-off sweep covers only MINE-Reg. Equal training budgets do not establish comparable privacy-utility operating points. Without baseline sweeps or comparisons at matched privacy or utility, the reported advantage could reflect regularization strength rather than the proposed mechanism.
Questions for authors
- What prevents a tighter additive bound under the assumptions used in Appendix A.1, and what result establishes a positional effect beyond that bound?
- Which dataset produced each comparison in Table 7, and do corrected per-seed PARI calculations preserve the reported ranking?
- What are the sensitive classes, how are they sampled and assigned to agents, and what attacker architecture, class frequencies, and split sizes underlie the final-output evaluation?
- Does MINE-Reg retain its advantage when the alternative interventions are tuned on the same validation split and compared at matched benign success or attack success?
2 comments
W3 / Q3: Sensitive Variables and Final-Output Adversarial Evaluation
Sensitive information. Our framework protects agent-local sensitive variables rather than assuming that all sensitive information corresponds to a single shared secret. Appendix A.3.4 specifies private-prompt embeddings, routing states, and policy indicators for MedQA/FinQA, while PrivacyLens uses protected seed attributes such as data subject, recipient, and transmission principle.
The attack objective is to predict agent-local sensitive attributes from the information exposed by the pipeline. In the final-output setting, the attacker observes only and predicts the protected attributes; it does not access the intermediate agent representations .
Independent adversarial evaluation. Section 6.5 (Table 3) evaluates this threat model on MedQA/Qwen-4B. Agent parameters are frozen, and the attacker is trained independently using a disjoint attack-training split.
Table 1: Final-output adversarial inference on MedQA/Qwen-4B (Table 3).
| Agents | Method | Macro Acc. | Macro F1 | BS |
|---|---|---|---|---|
| 3 | Base | 0.72 | 0.70 | 0.92 |
| 3 | MINE-Reg | 0.30 | 0.29 | 0.85 |
| 5 | Base | 0.78 | 0.76 | 0.88 |
| 5 | MINE-Reg | 0.32 | 0.31 | 0.81 |
MINE-Reg reduces attacker macro accuracy by and in the three-agent and five-agent settings, corresponding to relative reductions of and , respectively. Benign success remains and .
This evaluation is independent of the MINE critics optimized during training and directly tests sensitive-attribute recoverability from the final output.
The evaluation uses a four-class classification setting with a uniform-guessing reference of . The precise class mapping, class frequencies, attacker configuration, and split sizes are necessary to reproduce the attack protocol and assess class imbalance. We distinguish the stated uniform-guessing reference from a class-prior baseline and do not interpret reduced attacker accuracy as a certified MI upper bound.
W4 / Q4: Fairness of Privacy–Utility Comparisons
The alternative interventions in Appendix A.5.4 include Gaussian perturbation of intermediate representations (DP-style Noise) and an agent-local distribution-flattening privacy regularizer (PR). They use the same backbone and nominal training-budget settings as MINE-Reg.
The reviewer asks whether the advantage persists at comparable privacy or utility. The reported LLaMA-3B FinQA results provide informative operating-point comparisons:
Table 2: Privacy–utility comparison on FinQA using LLaMA-3B (Appendix Table 7; corrected dataset label).
| Agents | Method | MI$_{\mathrm{avg}}\downarrow$ | SB | BS |
|---|---|---|---|---|
| 2 | Base | 0.69 | 0.31 | 0.85 |
| DP | 0.18 | 0.68 | 0.52 | |
| PR | 0.36 | 0.57 | 0.74 | |
| MINE-Reg | 0.21 | 0.69 | 0.77 | |
| 3 | Base | 0.87 | 0.26 | 0.82 |
| DP | 0.22 | 0.64 | 0.48 | |
| PR | 0.48 | 0.52 | 0.72 | |
| MINE-Reg | 0.26 | 0.64 | 0.75 |
Comparison with DP-style Noise. With two agents, MINE-Reg achieves SB versus for DP-style Noise, while preserving substantially higher benign success ( versus ). With three agents, both methods achieve SB , but MINE-Reg retains BS compared with for DP-style Noise. These correspond to - and -percentage-point improvements in benign success at the reported equal-SB operating points.
Comparison with PR. For two agents, MINE-Reg achieves higher sensitive blocking ( versus ) and benign success ( versus ), with lower estimated MI ( versus ). The same pattern holds for three agents: SB improves from to , BS from to , and estimated MI decreases from to .
Thus, the reported results show a favorable privacy–utility balance that is not attributable solely to the composite PARI metric. In particular, MINE-Reg achieves stronger utility than DP-style Noise at equal reported sensitive blocking, and improves both blocking and utility relative to PR at the evaluated operating points. These comparisons do not establish dominance over independently optimized privacy–utility frontiers.
We thank the reviewer for the detailed assessment and for recognizing the relevance of our final-output adversarial evaluation. We clarify the theoretical interpretation, dataset labeling, attack protocol, and privacy–utility comparisons below.
W1 / Q1: Composition Bound and Position-Dependent Leakage
Theorem 1 concerns the relationship between agent-local leakage constraints and final-output privacy risk. The central distinction is between marginal leakage I(Oi;Si) and joint leakage I(ON;S1:N), which need not be characterized by the same information quantities.
As explained in our response to Reviewer ZwPM (W2/Q2), the direct additive recurrence requires additional conditional-independence conditions. Since each agent processes (Oi−1,Di,Si), conditional data processing gives
I(Oi;S<i∣Si)≤I(Oi−1,Di;S<i∣Si),
which cannot generally be reduced to an expression involving Oi−1 alone.
More precisely, defining Gi=I(Oi;S1:i), we obtain
Gi≤τi+Gi−1+δi+κi,
where
δi=I(S<i;Di∣Oi−1,Si),κi=I(S<i;Si∣Oi−1).
These terms capture dependencies involving stage-specific inputs and upstream representations. Under the sufficient independence conditions detailed in Reviewer ZwPM (W2/Q2), both terms vanish, yielding the tighter additive bound.
Under those conditions, the exponentially weighted expression in Theorem 1 remains a valid conservative upper bound. Its coefficients arise from the recursive bounding procedure and are not claimed to be tight lower bounds on positional leakage. Accordingly, Proposition 1 describes sensitivity of this upper bound to individual leakage budgets, rather than proving that earlier agents necessarily leak more information.
The empirical motivation remains distinct: intermediate representations can retain dependencies involving multiple agent-local secrets, and their final-output recoverability must be evaluated at the system level.
W2 / Q2: Table 7 Dataset Labeling and PARI Consistency
Dataset clarification. We thank the reviewer for identifying the labeling error. Appendix Table 7 reports alternative privacy-intervention comparisons on FinQA, not MedQA. The table caption and corresponding dataset references in Appendix A.5.4 were incorrectly labeled. Thus, the LLaMA-3B Base and MINE-Reg results in Table 7 correctly match FinQA Table 2 for both two-agent and three-agent settings. We will correct these labels in the revised PDF; the numerical entries in Tables 2 and 7 remain unchanged.
PARI Consistency. We thank the reviewer for raising this concern. The apparent discrepancy arises from reconstructing a nonlinear composite metric using component values reported to only two decimal places. Appendix A.4.6 defines
PARI=0.3,BS+0.1,OT+0.5(1−MIbaselineMIavg)+0.1,RT,
where RT=1.
Consider the FinQA/LLaMA-3B three-agent MINE-Reg configuration, for which Table 2 reports MI =0.26, baseline MI =0.87, BS =0.75, and PARI =0.781. Substituting these rounded values and setting OT =1 produces PARI ≈0.7756. However, the displayed precision does not uniquely determine the underlying component values.
Specifically, the reported entries correspond to the rounding intervals
MIavgMIbaselineBS∈[0.255,0.265),∈[0.865,0.875),∈[0.745,0.755).
These intervals admit PARI values up to approximately 0.7808 when OT =1, which round to 0.781. Thus, the reported PARI is mathematically compatible with the displayed component precision; substituting rounded table entries is not a valid test of exact numerical equality.
The distinction is particularly relevant because privacy integrity contains a ratio, so its value must be evaluated using the corresponding underlying MI quantities. Furthermore, averaging per-run ratios is generally not equivalent to taking the ratio of averaged MI values.
PARI serves as a composite summary, while the primary privacy–utility conclusions are independently supported by the reported MI, sensitive-blocking, and benign-success metrics. In particular, the comparisons in Appendix Table 7 can be evaluated directly through these primitive metrics without relying on the composite index.