acceptodds
Under review as a conference paper at ICLR 2027

The Behavioral Anatomy of Multi-Turn Jailbreaks: Critical Turns, Context Dependence, and Self-Conditioning

Abstract

Multi-turn jailbreak evaluations usually score transcripts the target model never produced: the schedule is flattened into one prompt, or source replies are teacher-forced into the history. Both remove the dependency that defines the setting — the model reads and continues its own prior output. We replay the canonical interaction instead: one local generation call per source user turn, each output appended verbatim, with per-turn state capture and fresh labelling. On the development split of a multi-turn attack corpus, replayed against one open-weight model google/gemma-2-9b-it, we complete 1916 generation calls over 480 units and label 1793 outputs with a declared evaluator panel — panel adjudications, not human ground truth. Three preregistered controlled branch experiments on one shared onset cohort then show: replacing the user turn immediately before first compliance onset removes downstream compliance (specific effect 0.500, 80% CI lower bound 0.364, permutation p = 0.0004, 22 source goals), while deletion and reordering controls do not; no shorter retained context reproduces onset compliance (best 0.286 against a 0.70 floor); and replacing the model's own prior assistant output collapses downstream compliance (0.656, 80% CI lower bound 0.500). We report the negatives with equal weight: three attempts at a pre-generation internal readout produced no validated one, so patching was not justified and was not run; neither sparse-autoencoder width qualified as an instrument; and a memory-policy experiment remains unresolved behind a label-calibration limit. The result is a development behavioral signature on a single model, not a mechanism, and the three angles share one cohort rather than replicating independently.

Primary area:
alignment, fairness, safety, privacy, and societal considerations
Keywords:
multi-turn jailbreaks, conversational safety, behavioral evaluation, context dependence, self-conditioning, mechanistic interpretability

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_u1DGOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

The paper replays multi-turn jailbreaks to ensure that the model conditions on its own earlier replies, and it uses preregistered branch interventions on a single open-weight model with outcomes labelled by a panel of LLM judges to show that compliance depends on the user turn just before onset, on the full conversation history, and on the prior output of the model itself. Attempts to locate an internal mechanism using probes and sparse autoencoders did not succeed.

Critical strengths

The paper replays multi-turn attacks so that the model conditions on its own earlier replies, and it builds a clean branch-intervention protocol on this foundation. The protocol uses equal-token replacements, matched controls, and a sham branch that must reproduce the original run. This protocol is the most reusable contribution of the work.

Critical weaknesses

The paper is only tested on the Gemma 9B model and on the development split only. The paper also reveals that LLM-panel labels could not be reproduced by a second panel (κ = 0.0229), which makes it more unreliable. I think the results could not be shown to extend beyond this model and the dataset.

These onsets were selected precisely because they were labelled compliant. Any change to the conversation, even a change that preserves the content, is therefore likely to eliminate that rare compliance, whereas non-compliant controls rarely change in the opposite direction. Reordering the pre-onset turn alone yields a rate of 0.409 compared with 0.500 for the main condition, and Section 9 does not attempt a change that preserves the content. The experiments therefore cannot establish whether the content matters or whether rare compliant outputs are simply fragile, which means that the headline effects cannot yet be interpreted as causal.

Also the paper has a clear trace of using AI autoresearch.

Questions for authors

Can you extend to more models?

Can you try a different dataset?

Can you add a content-preserving control, such as paraphrasing the pre-onset turn or the model's reply?

How often does an onset stay compliant if you regenerate it with a different seed and no intervention?

Which public multi-turn attack corpus did you use?

Reviewer_KAcyOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
1
Soundness
1
Read review
Paper summary

Rather than proposing new jailbreak attacks or developing defenses, this paper treats "multi-turn jailbreaks" like a serialized drama, replaying the interaction round by round to conduct controlled experiments. The goal is to pinpoint exactly which turn, which segment of the conversation history, or which of the model's own previous statements steered it off course. Titled "Behavioral Anatomy," the paper focuses exclusively on dissecting the model's behavior, making no claims about having deciphered its internal mechanisms.

Critical strengths
  1. The motivation of this paper quite interesting, and the analysis of multi-turn attacks is also a topic worth exploring.
  2. The negative results are a real contribution.
Critical weaknesses
  1. My biggest worry is external validity. The study uses one target model, one development split, and 22 source goals. The onset cohort is small. The paper itself says this is a development behavioral signature, so I will not accuse the authors of overclaiming. Still, I cannot generalize these effects to other models or to deployment settings.
  2. This paper is hard to read and contains many obscure terms. Could it be that the paper was generated with an over-reliance on LLM assistance?
  3. The measurement basis for the outcome variables is not robust; all labels are derived from an LLM panel rather than human annotation.
Questions for authors

Please refer to Critical Weaknesses

Reviewer_fKS8OpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

The paper studies multi-turn jailbreaks strategies in a realistic setting (without flattening the conversation)

Critical strengths
  • I am very sorry but I don't find any strength in this paper
Critical weaknesses
  • the paper lacks of academic writing and appears as a draft with residual notes (e.g., instructions in results section)
  • the research does not properly engage with previous literature and results are not compared against existing baselines, methods, approaches.
Questions for authors

there are no questions whose answers could change my assessment, sorry.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.