- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper investigates differences in safety refusal between conversational and tool-calling settings. Across eight open-weight models, the authors find that models refuse harmful requests less frequently when tools are available. Through representation analysis and intervention experiments, they further argue that tool mediation preserves information about harmfulness but changes how that information is translated into refusal. The research question is important, and the paper provides substantial evidence of a behavioral gap. However, I find the evidence for the proposed mechanistic explanation less conclusive. My main concerns involve the comparability of refusal measurements, the causal connection between the identified internal components and refusal behavior, and the controls used to isolate the effect of tool exposure.
Critical strengths
- Important and practically relevant research question. As models increasingly perform actions through tools, understanding whether conversational safety transfers to tool use is valuable.
- Broad behavioral evaluation. The study covers eight models and compares different presentations of matched intents. The consistent reduction in tool-side refusal suggests that the phenomenon deserves attention.
- Useful controls and supplementary analyses. Multiple directive paraphrases, cross-fitting by intent, matched random-neuron controls, multi-turn evaluation, and the PrivacyLens experiments strengthen the study. The authors also acknowledge capability limitations and exceptions across architectures.
Critical weaknesses
- Refusal is measured differently across the two settings
According to Section 4 and Appendix B.2, conversational refusal is primarily assessed by an LLM judge, whereas tool-side compliance is primarily determined by whether the model calls a target function. These procedures do not necessarily measure the same behavior. For example, the absence of a target call may reflect refusal, insufficient tool-use capability, or malformed output. Conversely, a target-function call should be evaluated in context to establish whether it attempts a harmful action. The measured gap may therefore combine differences in safety behavior with differences in execution capability and output format. The capability-related analysis and multi-turn evaluation are helpful. However, because refusal labels also inform the subsequent representation and threshold analyses, a common evaluation standard is needed. I would like to see both settings assessed using the same categories: explicit safety refusal, harmful assistance, capability failure, and formatting failure, with harmful tool-call attempts reported separately.
- Changes in internal signals do not yet establish changes in the refusal mechanism
In Section 5.1 and Tables 15–16, the main outcome of the neuron-ablation experiments is the change in the final-layer harmfulness projection. This establishes that the selected neurons contribute to the measured representation, but it does not directly establish that they determine whether the model refuses a harmful request. An alternative explanation remains possible: tool mediation changes how information is encoded, producing different neuron contributions and projection values, while the computation responsible for refusal remains substantially shared. Appendix C.7 does include directional interventions that measure refusal behavior. However, these interventions differ from the neuron sets identified earlier, leaving a gap between locating channel-related components and demonstrating their behavioral role. The mechanistic claim would be stronger if ablating the identified neurons selectively changed refusal in the corresponding channel while preserving benign tool-use capability. A restoration or activation-patching experiment would provide additional support, although a clear behavioral validation of the existing neuron sets would already be valuable.
- The effects of tool availability and execution instructions are not fully separated
The system prompt for tool exposure₁ in Table 7 already instructs the model to issue tool calls when appropriate. Results under this condition therefore provide limited support for the claim that tool exposure triggers the shift before the model is instructed to act. The tool exposure₂ condition removes this explicit instruction, which is a useful improvement. However, the paper lacks a direct no-tool control using exactly the same system prompt as exposure₂. Moreover, the main representational analysis in Figure 3 is based on exposure₁. A more controlled comparison would hold the system prompt and user instruction fixed while changing only tool availability. Repeating the key behavioral and representational analyses under this comparison would clarify how much of the observed shift can be attributed to tool exposure itself.
Questions for authors
-
Does the main refusal gap persist when both settings are evaluated under a common annotation scheme distinguishing safety refusal, harmful assistance, capability failure, and formatting failure? How are tool-side outputs without a target call classified?
-
How does ablating the identified channel-related neuron sets affect actual refusal and harmful tool-call rates? Are these effects channel-selective, and do they occur while benign tool-use capability remains intact?
-
What happens in a no-tool condition using exactly the same system prompt as tool exposure₂? Can the representational findings in Figure 3 be reproduced under this more controlled comparison?
3 comments
Reviewer Tib9 (2/3)
Behavioral effects of neuron experiments. We agree that testing the identified neurons behaviorally would complement the representational results in the paper. We conducted two additional experiments to further examine this: (1) an ablation, and (2), a restoration where we restore activations to the corresponding conversational mean values. In both cases, we intervene on the identified tool-selective and conversation-selective populations (K=8000, four late MLP layers) and compare the refusal rates in the tool channel against a random control set. We present the results for the smaller models in the following table, where we show refusal rates before and after intervention (on all intents, including those with execution hints) and the resulting benign capability after the intervention. We omit the ablation for Llama-3.1-8B as its baseline refusal rate is 0.05.
| Model | Baseline refusal | After ablation | After restoration | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Random | Tool-sel. | Conv-sel. | Benign | Random | Tool-sel. | Conv-sel. | Benign | ||||
| Llama-3.1-8B | 0.05 | - | - | - | - | 0.05 | 0.08 | 0.16 | 0.98 / 0.98 | ||
| Qwen3-8B | 0.28 | 0.27 | 0.20 | 0.22 | 0.99 / 0.99 | 0.27 | 0.35 | 0.43 | 0.99 / 0.98 | ||
| Ministral-3-8B | 0.34 | 0.34 | 0.30 | 0.36 | 1.00 / 1.00 | 0.37 | 0.38 | 0.53 | 1.00 / 0.99 | ||
| Gemma-4-12B | 0.63 | 0.63 | 0.15 | 0.64 | 0.34 / 0.95 | 0.64 | 0.92 | 0.64 | 0.46 / 0.97 |
Restoring the conversation-selective population increases refusal in all three 8B models. Conversely, ablating the tool-selective population reduces refusal in Qwen3-8B and Ministral-3-8B, and collapses benign capability in Gemma-4-12B (we also tested this model at K=800, where there is a 0.07 drop in refusal without capability degradation). Benign tool execution remains largely unaffected across interventions, except for tool-selective interventions in Gemma-4-12B, where there is a collapse in benign tool execution. Taken together, these results indicate that, at least in the smaller models, the channel-specific populations also play a role in tool-channel refusal, and that tool mediation does not simply weaken a single shared population, but instead relies partly on a different one.
Ablating the conversational channel leaves refusal unchanged, consistent with the robustness of conversational refusal in Appendix C.7 and Table 4. In the larger models, we do not observe major changes in refusal from these interventions. A fixed set of 8,000 neurons in four layers covers a smaller fraction of the network in these models, which limits the strength of the intervention. We will add these experiments in the appendix, summarise them in Sections 1 and 5.1, and state their scope explicitly in the Abstract and Introduction.
Response to Reviewer Tib9 (3/3)
Effect of execution instructions in refusal. We ran the requested control, considering the exact system prompt from tool-exp₂ with no tools attached. Across all eight models, refusal in this condition remains close or higher than the conversational level, and attaching tools reduces refusal in every model. This effect is comparable to that under tool-exp₁, suggesting that the explicit instruction to execute tool calls in the system prompt is not necessary for the behavioral shift.
| Model | Reference | tool-exp₂ prompt, no tools | tool-exp₂ prompt, with tools | tool-exp₂ | tool-exp₁ - agentic |
|---|---|---|---|---|---|
| Qwen3-8B | 0.971 | 0.993 | 0.787 | −0.206 | −0.245 |
| Qwen3-32B | 0.968 | 0.985 | 0.811 | −0.174 | −0.239 |
| Llama-3.1-8B | 0.941 | 0.986 | 0.551 | −0.435 | −0.568 |
| Llama-3.1-70B | 0.892 | 0.973 | 0.715 | −0.258 | −0.332 |
| Ministral-3-8B | 0.725 | 0.771 | 0.642 | −0.129 | −0.271 |
| Gemma-4-12B | 0.935 | 0.967 | 0.871 | −0.095 | −0.090 |
| Gemma-4-31B | 0.945 | 0.960 | 0.857 | −0.103 | −0.131 |
| gpt-oss-20b | 0.998 | 0.995 | 0.889 | −0.106 | −0.155 |
Representational results. We repeated the analyses from Figure 3 using tool-exp₂, with the results are shown in the table below. The findings are consistent with Figure 3 (despite tool-action using the tool-exp₁ prompt), and confirm that the representational findings also hold without the explicit instruction to execute tool calls in the system prompt.
| Model | Exposure fraction, tool-exp₁ (Figure 3) | Exposure fraction tool-exp₂ | Axis agreement ÷ ceiling, exposure₁ / exposure₂ | Transfer AUROC, exposure₁ / exposure₂ |
|---|---|---|---|---|
| Qwen3-8B | 0.76 | 0.65 | 0.87 / 0.85 | 0.98 / 0.98 |
| Qwen3-32B | 0.71 | 0.67 | 0.58 / 0.45 | 0.78 / 0.74 |
| Llama-3.1-8B | 0.86 | 0.82 | 0.87 / 0.84 | 0.99 / 0.99 |
| Llama-3.1-70B | 0.72 | 0.48 | 0.81 / 0.77 | 1.00 / 1.00 |
| Ministral-3-8B | 0.80 | 0.69 | 0.95 / 0.92 | 0.92 / 0.91 |
| Gemma-4-12B | 0.63 | 0.53 | 0.90 / 0.87 | 0.98 / 0.99 |
| Gemma-4-31B | 0.60 | 0.50 | 0.92 / 0.89 | 0.97 / 0.98 |
| gpt-oss-20b | 0.78 | 0.64 | 0.96 / 0.94 | 0.95 / 0.95 |
We additionally decompose the fractions relative to the matched no-tool condition (tool-exp₂ system prompt without tools) to further isolate the representation impact of tool availability. Attaching tools is the dominant effect to the representational shift in all models, although the agentic framing already shifts the representation by 4–26% in some. Overall, this indicates that exposure is the primary driver of the representational shift, while the agentic framing makes a smaller, model-dependent contribution.
| Model | tool-exp₂ prompt, no tools | tool-exp₂ |
|---|---|---|
| Qwen3-8B | −10% | 65% |
| Qwen3-32B | −7% | 67% |
| Llama-3.1-8B | −10% | 82% |
| Llama-3.1-70B | −20% | 48% |
| Ministral-3-8B | 4% | 69% |
| Gemma-4-12B | 26% | 53% |
| Gemma-4-31B | 15% | 50% |
| gpt-oss-20b | 26% | 64% |
We will add the behavioral and representational results in Appendix C.2. We will also state that the observed representational effects persist without the explicit instruction to execute tool calls in the system prompt, while noting the contribution of agentic prompt framing in some models.
Response to Reviewer Tib9 (1/3)
We thank the reviewer for taking the time to provide detailed feedback on our work. We appreciate the helpful suggestions and address each of the points raised below.
Common annotation scheme. We re-graded harmful intents using Qwen3.5-27B as an LLM judge, applying the same evaluation procedure to conversational and tool-mediated outputs. The refusal drop reported in Table 2 persists across all eight models and is close to the results obtained using our original grading.
We also extend the capability comparison in the paper to cover both conversational and tool-mediated outputs, distinguishing the categories requested, using both the original grading procedure used in the paper, as well as Qwen3.5-27B instructed to distinguish between those categories, with the following results:
Original grading, Conversational → Tool:
LLM judge, Conversational → Tool:
Results show in both procedures that the reduction in explicit safety refusals in the tool-mediated setting is accompanied by a corresponding increase in harmful assistance, while capability and formatting failures remain negligible and cannot account for the observed refusal drop. We will extend Appendix B.2 to include these additional evaluations and the corresponding categorical taxonomy.
Tool-side outputs without a function call. We classify those instances as refusal if they contain an explicit refusal statement, and otherwise as a format failure, which we exclude for the analysis. Counting all indeterminate outputs as refusals raises tool-side refusal rates by at most 0.06, still below conversational refusal rates for every model. Moreover, the new LLM judge evaluation classifies 89–100% of these outputs as harmful assistance for all models except Ministral-3-8B (54%), so their exclusion might understate the refusal gap. We will clarify this rule in the revised manuscript.