Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2605.00267

Jailbroken Frontier Models Retain Their Capabilities

Daniel Zhu·Zihan Wang·Xuchan Bao·Jerry Wei
View original source ↗Read source document ↗Export metadata ↓

Abstract

As language model safeguards become more robust, attackers are pushed toward developing increasingly complex jailbreaks. Prior work has found that this complexity imposes a "jailbreak tax" that degrades the target model's task performance. We show that this tax scales inversely with model capability and that the most advanced jailbreaks effectively yield no reduction in model capabilities. Evaluating 28 jailbreaks on five benchmarks across Claude models ranging in capability from Haiku 4.5 to Opus 4.6, we find Haiku 4.5 loses an average of 33.1% on benchmark performance when jailbroken, while Opus 4.6 at max thinking effort loses only 7.7%. We also observe that across all models, reasoning-heavy tasks display considerably more degradation than knowledge-recall tasks. Finally, Boundary Point Jailbreaking, currently the strongest jailbreak against deployed classifiers, achieves near-perfect classifier evasion with near-zero degradation across safeguarded models. We recommend that safety cases for frontier models should not rely on a meaningful capability degradation from jailbreaks.

Explore related subjects

cs.LGcs.AIcs.CR

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Daniel Zhu, Zihan Wang, Xuchan Bao, Jerry Wei. 2026-05-04. Jailbroken Frontier Models Retain Their Capabilities. https://arxiv.org/abs/2605.00267

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Incentivizing Honesty among Competitors in Collaborative Learning and Optimization

Collaborative learning techniques have the potential to enable training machine learning models that are superior to models trained on a single entity's data. However, in many cases, potential participants in such collaborative schemes are competitors on a downstream task, such as firms that each aim to attract customers by providing the best recommendations. This can incentivize dishonest updates that damage other participants' models, potentially undermining the benefits of collaboration. In this work, we formulate a game that models such interactions and study two learning tasks within this framework: single-round mean estimation and multi-round SGD on strongly-convex objectives. For a natural class of player actions, we show that rational clients are incentivized to strongly manipulate their updates, preventing learning. We then propose mechanisms that incentivize honest communication and ensure learning quality comparable to full cooperation. Lastly, we empirically demonstrate the effectiveness of our incentive scheme on a standard non-convex federated learning benchmark. Our work shows that explicitly modeling the incentives and actions of dishonest clients, rather than assuming them malicious, can enable strong robustness guarantees for collaborative learning.

cs.LG↗
arXiv2026

Hierarchical Deep Counterfactual Regret Minimization

Imperfect Information Games (IIGs) are used to model games under uncertainty or lack complete information. Counterfactual Regret Minimization (CFR) is one of the most successful families of algorithms for IIGs. The integration of skill-based strategy learning with CFR could potentially mirror more human-like decision-making and improve learning on complex IIGs. It enables the learning of a hierarchical strategy, wherein low-level components represent skills for solving subgames and the high-level component manages the transition between skills. In this paper, we introduce the first hierarchical version of Deep CFR (HDCFR), an innovative method that boosts learning efficiency in tasks involving extensively large state spaces and deep game trees. Notably, HDCFR enables learning with predefined (human) expertise and extracting skills transferable to similar tasks. We first present the algorithm and establish its theory in a tabular setting, including hierarchical CFR update rules and a variance-reduced Monte Carlo sampling extension for the model-free setting, where backtracking is infeasible. We then extend HDCFR to large-scale tasks via deep learning objectives that match the tabular targets under exact function fitting. Code: https://anonymous.4open.science/r/HDCFR_RUN-677B.

cs.LG↗
arXiv2026

Achieving Linear Speedup with ProxSkip in Distributed Stochastic Optimization

The ProxSkip algorithm for distributed optimization is gaining increasing attention due to its effectiveness in reducing communication. However, existing analyses of ProxSkip are limited to the strongly convex setting and fail to achieve linear speedup with respect to the number of nodes. Key questions regarding its behavior in the non-convex setting and the achievability of linear speedup remain open. In this paper, we revisit decentralized ProxSkip and answer these questions affirmatively. We provide a unified convergence analysis for stochastic non-convex, convex, and strongly convex problems, revealing how gradient noise, local updates, network connectivity, and data heterogeneity jointly determine the convergence behavior. To the best of our knowledge, this is the first analysis showing that decentralized ProxSkip achieves linear speedup in the number of nodes under stochastic gradients. Moreover, our results demonstrate that local updates can effectively reduce communication frequency and improve communication efficiency.

cs.LG↗