Large (vision-)language models exhibit remarkable capability but remain highly susceptible to jailbreaking. Existing safety training approaches aim to have the model learn a refusal boundary between safe and unsafe, based on the user's intent. It has been found that this binary training regime often leads to brittleness, since the user intent cannot reliably be evaluated, especially if the attacker obfuscates their intent, and also makes the system seem unhelpful. In response, frontier models, such as GPT-5, have shifted from refusal-based safeguards to safe completion, that aims to maximize helpfulness while obeying safety constraints. However, safe completion could be exploited when a user pretends their intention is benign. Specifically, this intent inversion would be effective in multi-turn conversation, where the attacker has multiple opportunities to reinforce their deceptively benign intent. In this work, we introduce a novel multi-turn jailbreaking method that exploits this vulnerability. Our approach gradually builds conversational trust by simulating benign-seeming intentions and by exploiting the consistency property of the model, ultimately guiding the target model toward harmful, detailed outputs. Most crucially, our approach also uncovered an additional class of model vulnerability that we call para-jailbreaking that has been unnoticed up to now. Para-jailbreaking describes the situation where the model may not reveal harmful direct reply to the attack query, however the information that it reveals is nevertheless harmful. Our contributions are threefold. First, it achieves high success rates against frontier models including GPT-5-thinking and Claude-Sonnet-4.5. Second, our approach revealed and addressed para-jailbreaking harmful output. Third, experiments on multimodal VLM models showed that our approach outperformed state-of-the-art models.

Multi-agent large language model (LLM) architectures increasingly rely on response-level aggregation, such as Majority Voting (MAJ), to raise reasoning ceilings. However, in open environments, agents are highly susceptible to stealthy contextual corruption, such as targeted prompt injections. We reveal a critical structural vulnerability in current multi-agent systems: response-level aggregation collapses when corrupted agents form a local majority. Because voting aggregates fully-formed conclusions, it is blind to flawed intermediate logic. To overcome this systematic limitation, we propose the Token-Level Round-Robin (RR) Collaboration, where agents sequentially interleave generation within a shared auto-regressive context. We formalize this process as a discrete-time dynamical system, proving that token-level interleaving transitions aggregation from a brittle counting of final votes (a linear sum) to a dynamic, interwoven chain of logic (a non-linear operator product). Through this theoretical lens, we prove that the honest model's restorative pull can overpower adversarial corruptions, even when corrupted agents form a majority. We conduct an exhaustive empirical evaluation across diverse reasoning benchmarks and demonstrate that while MAJ collapses when corrupted agents reach a majority, RR maintains robust accuracy well beyond this critical threshold.

The increasing adaptation of vision models across domains, such as satellite imagery and medical scans, has raised an emerging privacy risk: models may inadvertently retain and leak sensitive source-domain specific information in the target domain. This creates a compelling use case for machine unlearning to protect the privacy of sensitive source-domain data. Among adaptation techniques, source-free domain adaptation (SFDA) calls for an urgent need for machine unlearning (MU), where the source data itself is protected, yet the source model exposed during adaptation encodes its influence. Our experiments reveal that existing SFDA methods exhibit strong zero-shot performance on source-exclusive classes in the target domain, indicating they inadvertently leak knowledge of these classes into the target domain, even when they are not represented in the target data. We identify and address this risk by proposing an MU setting called SCADA-UL: Unlearning Source-exclusive ClAsses in Domain Adaptation. Existing MU methods do not address this setting as they are not designed to handle data distribution shifts. We propose a new unlearning method, where an adversarially generated forget class sample is unlearned by the model during the domain adaptation process using a novel rescaled labeling strategy and adversarial optimization. We also extend our study to two variants: a continual version of this problem setting and to one where the specific source classes to be forgotten may be unknown. Alongside theoretical interpretations, our comprehensive empirical results show that our method consistently outperforms baselines in the proposed setting while achieving retraining-level unlearning performance on benchmark datasets. Our code is available at https://github.com/D-Arnav/SCADA

Deep neural networks (DNNs) have achieved remarkable success in computer vision but remain highly vulnerable to adversarial attacks. Among them, camouflage attacks manipulate an object's visible appearance to deceive detectors while remaining stealthy to humans. In this paper, we propose a new framework that formulates vehicle camouflage attacks as a conditional image-editing problem. Specifically, we explore both image-level and scene-level camouflage generation strategies, and fine-tune a ControlNet to synthesize camouflaged vehicles directly on real images. We design a unified objective that jointly enforces vehicle structural fidelity, style consistency, and adversarial effectiveness. Extensive experiments on the COCO and LINZ datasets show that our method achieves significantly stronger attack effectiveness, leading to more than 38% AP50 decrease, while better preserving vehicle structure and improving human-perceived stealthiness compared to existing approaches. Furthermore, our framework generalizes effectively to unseen black-box detectors and exhibits promising transferability to the physical world. Project page is available at https://humansensinglab.github.io/CtrlCamo

LLM based agents are increasingly deployed in high stakes settings where they process external data sources such as emails, documents, and code repositories. This creates exposure to indirect prompt injection attacks, where adversarial instructions embedded in external content manipulate agent behavior without user awareness. A critical but underexplored dimension of this threat is concealment: since users tend to observe only an agent's final response, an attack can conceal its existence by presenting no clue of compromise in the final user facing response while successfully executing harmful actions. This leaves users unaware of the manipulation and likely to accept harmful outcomes as legitimate. We present findings from a large scale public red teaming competition evaluating this dual objective across three agent settings: tool calling, coding, and computer use. The competition attracted 464 participants who submitted 272000 attack attempts against 13 frontier models, yielding 8648 successful attacks across 41 scenarios. All models proved vulnerable, with attack success rates ranging from 0.5% (Claude Opus 4.5) to 8.5% (Gemini 2.5 Pro). We identify universal attack strategies that transfer across 21 of 41 behaviors and multiple model families, suggesting fundamental weaknesses in instruction following architectures. Capability and robustness showed weak correlation, with Gemini 2.5 Pro exhibiting both high capability and high vulnerability. To address benchmark saturation and obsoleteness, we will endeavor to deliver quarterly updates through continued red teaming competitions. We open source the competition environment for use in evaluations, along with 95 successful attacks against Qwen that did not transfer to any closed source model. We share model-specific attack data with respective frontier labs and the full dataset with the UK AISI and US CAISI to support robustness research.

Multi-modal Large Language Models (MLLMs) have achieved remarkable performance across a wide range of visual reasoning tasks, yet their vulnerability to safety risks remains a pressing concern. While prior research primarily focuses on jailbreak defenses that detect and refuse explicitly unsafe inputs, such approaches often overlook contextual safety, which requires models to distinguish subtle contextual differences between scenarios that may appear similar but diverge significantly in safety intent. In this work, we present MM-SafetyBench++, a carefully curated benchmark designed for contextual safety evaluation. Specifically, for each unsafe image-text pair, we construct a corresponding safe counterpart through minimal modifications that flip the user intent while preserving the underlying contextual meaning, enabling controlled evaluation of whether models can adapt their safety behaviors based on contextual understanding. Further, we introduce EchoSafe, a training-free framework that maintains a self-reflective memory bank to accumulate and retrieve safety insights from prior interactions. By integrating relevant past experiences into current prompts, EchoSafe enables context-aware reasoning and continual evolution of safety behavior during inference. Extensive experiments on various multi-modal safety benchmarks demonstrate that EchoSafe consistently achieves superior performance, establishing a strong baseline for advancing contextual safety in MLLMs. All benchmark data and code are available at https://echosafe-mllm.github.io.

Keypoint detection underpins many vision tasks, including pose estimation, viewpoint recovery, and 3D reconstruction, yet modern neural models remain vulnerable to small input perturbations. Despite its importance, formal robustness verification for keypoint detectors is largely unexplored due to high-dimensional inputs and continuous coordinate outputs. We propose the first coupled robustness verification framework for heatmap-based keypoint detectors that bounds the joint deviation across all keypoints, capturing their interdependencies and downstream task requirements. Unlike prior decoupled, classification-style approaches that verify each keypoint independently and yield conservative guarantees, our method verifies collective behavior. We formulate verification as a falsification problem using a mixed-integer linear program (MILP) that combines reachable heatmap sets with a polytope encoding joint deviation constraints. Infeasibility certifies robustness, while feasibility provides counterexamples, and we prove the method is sound: if it certifies the model as robust, then the keypoint detection model is guaranteed to be robust. Experiments show that our coupled approach achieves high verified rates and remains effective under strict error thresholds where decoupled methods fail.

Pruning-based unlearning has recently emerged as a fast, training-free, and data-independent approach to remove undesired concepts from diffusion models. It promises high efficiency and robustness, offering an attractive alternative to traditional fine-tuning or editing-based unlearning. However, in this paper we uncover a hidden danger behind this promising paradigm. We find that the locations of pruned weights, typically set to zero during unlearning, can act as side-channel signals that leak critical information about the erased concepts. To verify this vulnerability, we design a novel attack framework capable of reviving erased concepts from pruned diffusion models in a fully data-free and training-free manner. Our experiments confirm that pruning-based unlearning is not inherently secure, as erased concepts can be effectively revived without any additional data or retraining. Extensive experiments on diffusion-based unlearning based on concept related weights lead to the conclusion: once the critical concept-related weights in diffusion models are identified, our method can effectively recover the original concept regardless of how the weights are manipulated. Finally, we explore potential defense strategies and advocate safer pruning mechanisms that conceal pruning locations while preserving unlearning effectiveness, providing practical insights for designing more secure pruning-based unlearning frameworks.

Modern language models (LM) are trained on large scrapes of the Web, containing millions of personal information (PI) instances, many of which LMs memorize, increasing privacy risks. In this work, we develop the regexes and rules (R&R) detector suite to detect email addresses, phone numbers, and IP addresses, which outperforms the best regex-based PI detectors. On a manually curated set of 483 instances of PI, we measure memorization: finding that 13.6% are parroted verbatim by the Pythia-6.9b model, i.e., when the model is prompted with the tokens that precede the PI in the original document, greedy decoding generates the entire PI span exactly. We expand this analysis to study models of varying sizes (160M-6.9B) and pretraining time steps (70k-143k iterations) in the Pythia model suite and find that both model size and amount of pretraining are positively correlated with memorization. Even the smallest model, Pythia-160m, parrots 2.7% of the instances exactly. Consequently, we strongly recommend that pretraining datasets be aggressively filtered and anonymized to minimize PI parroting.

We report an exploratory red-teaming study of autonomous language-model-powered agents deployed in a live laboratory environment with persistent memory, email accounts, Discord access, file systems, and shell execution. Over a two-week period, twenty AI researchers interacted with the agents under benign and adversarial conditions. Focusing on failures emerging from the integration of language models with autonomy, tool use, and multi-party communication, we document eleven representative case studies. Observed behaviors include unauthorized compliance with non-owners, disclosure of sensitive information, execution of destructive system-level actions, denial-of-service conditions, uncontrolled resource consumption, identity spoofing vulnerabilities, cross-agent propagation of unsafe practices, and partial system takeover. In several cases, agents reported task completion while the underlying system state contradicted those reports. We also report on some of the failed attempts. Our findings establish the existence of security-, privacy-, and governance-relevant vulnerabilities in realistic deployment settings. These behaviors raise unresolved questions regarding accountability, delegated authority, and responsibility for downstream harms, and warrant urgent attention from legal scholars, policymakers, and researchers across disciplines. This report serves as an initial empirical contribution to that broader conversation.

Evaluation and alignment pipelines for large language models increasingly rely on LLM-based judges, whose behavior is guided by natural-language rubrics and validated on benchmarks. We identify a previously under-recognized vulnerability in this workflow, which we term Rubric-Induced Preference Drift (RIPD). Even when rubric edits pass benchmark validation, they can still produce systematic and directional shifts in a judge's preferences on target domains. Because rubrics serve as a high-level decision interface, such drift can emerge from seemingly natural, criterion-preserving edits and remain difficult to detect through aggregate benchmark metrics or limited spot-checking. We further show this vulnerability can be exploited through rubric-based preference attacks, in which benchmark-compliant rubric edits steer judgments away from a fixed human or trusted reference on target domains, systematically inducing RIPD and reducing target-domain accuracy up to 9.5% (helpfulness) and 27.9% (harmlessness). When these judgments are used to generate preference labels for downstream post-training, the induced bias propagates through alignment pipelines and becomes internalized in trained policies. This leads to persistent and systematic drift in model behavior. Overall, our findings highlight evaluation rubrics as a sensitive and manipulable control interface, revealing a system-level alignment risk that extends beyond evaluator reliability alone. The code is available at: https://github.com/ZDCSlab/Rubrics-as-an-Attack-Surface. Warning: Certain sections may contain potentially harmful content that may not be appropriate for all readers.

Large reasoning models with reasoning capabilities achieve state-of-the-art performance on complex tasks, but their robustness under multi-turn adversarial pressure remains underexplored. We evaluate nine frontier reasoning models under adversarial attacks. Our findings reveal that reasoning confers meaningful but incomplete robustness: most reasoning models studied significantly outperform instruction-tuned baselines, yet all exhibit distinct vulnerability profiles, with misleading suggestions universally effective and social pressure showing model-specific efficacy. Through trajectory analysis, we identify five failure modes (Self-Doubt, Social Conformity, Suggestion Hijacking, Emotional Susceptibility, and Reasoning Fatigue) with the first two accounting for 50% of failures. We further demonstrate that Confidence-Aware Response Generation (CARG), effective for standard LLMs, fails for reasoning models due to overconfidence induced by extended reasoning traces; counterintuitively, random confidence embedding outperforms targeted extraction. Our results highlight that reasoning capabilities do not automatically confer adversarial robustness and that confidence-based defenses require fundamental redesign for reasoning models.

Model distillation enables efficient emulation of frontier large language models (LLMs), creating a need for robust mechanisms to detect when a third-party student model has trained on a teacher model's outputs. However, existing fingerprinting techniques that could be used to detect such distillation rely on heuristic perturbations that impose a steep trade-off between generation quality and fingerprinting strength, often requiring significant degradation of utility to ensure the fingerprint is effectively internalized by the student. We introduce antidistillation fingerprinting (ADFP), a principled approach that aligns the fingerprinting objective with the student's learning dynamics. Building upon the gradient-based framework of antidistillation sampling, ADFP utilizes a proxy model to identify and sample tokens that directly maximize the expected detectability of the fingerprint in the student after fine-tuning, rather than relying on the incidental absorption of the un-targeted biases of a more naive watermark. Experiments on GSM8K and OASST1 benchmarks demonstrate that ADFP achieves a significant Pareto improvement over state-of-the-art baselines, yielding stronger detection confidence with minimal impact on utility, even when the student model's architecture is unknown.

Lipschitz-based certification offers efficient, deterministic robustness guarantees but has struggled to scale in model size, training efficiency, and ImageNet performance. We introduce \emph{LipNeXt}, the first \emph{constraint-free} and \emph{convolution-free} 1-Lipschitz architecture for certified robustness. LipNeXt is built using two techniques: (1) a manifold optimization procedure that updates parameters directly on the orthogonal manifold and (2) a \emph{Spatial Shift Module} to model spatial pattern without convolutions. The full network uses orthogonal projections, spatial shifts, a simple 1-Lipschitz $β$-Abs nonlinearity, and $L_2$ spatial pooling to maintain tight Lipschitz control while enabling expressive feature mixing. Across CIFAR-10/100 and Tiny-ImageNet, LipNeXt achieves state-of-the-art clean and certified robust accuracy (CRA), and on ImageNet it scales to 1-2B large models, improving CRA over prior Lipschitz models (e.g., up to $+8\%$ at $\varepsilon{=}1$) while retaining efficient, stable low-precision training. These results demonstrate that Lipschitz-based certification can benefit from modern scaling trends without sacrificing determinism or efficiency.

Serverless computing has achieved widespread adoption, with over 70% of AWS organizations using serverless solutions [1]. Meanwhile, machine learning inference workloads increasingly migrate to Function-as-a-Service (FaaS) platforms for their scalability and cost-efficiency [2], [3], [4]. However, this convergence introduces critical security challenges, with recent reports showing a 220% increase in AI/ML vulnerabilities [5] and serverless computing's fragmented architecture raises new security concerns distinct from traditional cloud deployments [6], [7]. This paper presents the first comprehensive security analysis of machine learning workloads in serverless environments. We systematically characterize the attack surface across five categories: function-level vulnerabilities (cold start exploitation, dependency poisoning), model-specific threats (API-based extraction, adversarial inputs), infrastructure attacks (cross-function contamination, privilege escalation), supply chain risks (malicious layers, backdoored libraries), and IAM complexity (ephemeral nature, serverless functions). Through empirical assessments across AWS Lambda, Azure Functions, and Google Cloud Functions, we demonstrate real-world attack scenarios and quantify their security impact. We propose Serverless AI Shield (SAS), a multi-layered defense framework providing pre-deployment validation, runtime monitoring, and post-execution forensics. Our evaluation shows SAS achieves 94% detection rates while maintaining performance overhead below 9% for inference latency. We release an open-source security toolkit to enable practitioners to assess and harden their serverless AI deployments, advancing the field toward more resilient cloud-native machine learning systems.

The open-weight language model ecosystem is increasingly defined by model composition techniques (such as weight merging, speculative decoding, and vocabulary expansion) that remix capabilities from diverse sources. A critical prerequisite for applying these methods across different model families is tokenizer transplant, which aligns incompatible vocabularies to a shared embedding space. We demonstrate that this essential interoperability step introduces a supply-chain vulnerability: we engineer a single breaker token that is functionally inert in a donor model yet reliably reconstructs into a high-salience malicious feature after transplant into a base model. By exploiting the geometry of coefficient reuse, our attack sabotages the base model's generation while leaving the donor's utility statistically indistinguishable from nominal behavior. We formalize this as a dual-objective optimization problem and instantiate the attack using a sparse solver. Empirically, the attack is training-free and evades outlier detection, while demonstrating structural persistence against fine-tuning and weight merging, highlighting a hidden risk in the pipeline of modular AI composition. Code is available at https://github.com/xz-liu/tokenforge

This paper introduces Jailbreak-Zero, a novel red teaming methodology that shifts the paradigm of Large Language Model (LLM) safety evaluation from a constrained example-based approach to a more expansive and effective policy-based framework. By leveraging an attack LLM to generate a high volume of diverse adversarial prompts and then fine-tuning this attack model with a preference dataset, Jailbreak-Zero achieves Pareto optimality across the crucial objectives of policy coverage, attack strategy diversity, and prompt fidelity to real user inputs. The empirical evidence demonstrates the superiority of this method, showcasing significantly higher attack success rates against both open-source and proprietary models like GPT-40 and Claude 3.5 when compared to existing state-of-the-art techniques. Crucially, Jailbreak-Zero accomplishes this while producing human-readable and effective adversarial prompts with minimal need for human intervention, thereby presenting a more scalable and comprehensive solution for identifying and mitigating the safety vulnerabilities of LLMs.

Watermarking aims to embed hidden signals in generated text that can be reliably detected when given access to a secret key. Open-weight language models pose acute challenges for such watermarking schemes because the inference-time interventions that dominate contemporary approaches cannot be enforced once model weights are public. Existing watermaking techniques for open-weight models, such as the recently proposed GaussMark, typically rely on small modifications to model weights, which can yield signals detectable to those equipped with a secret key, but achieving detection power comparable to inference-time watermarks generally requires weight perturbations that noticeably reduce generation quality. We introduce MarkTune, a theoretically principled, on-policy fine-tuning framework that treats the GaussMark signal as a reward while simultaneously regularizing against degradation in text quality. We derive MarkTune as an improvement on GaussMark and demonstrate that MarkTune consistently improves the quality-detectability trade-off over GaussMark by steering finer-grained, watermark-aware weight updates within the model's representation space while preserving generation quality. Empirically, we show that MarkTune pushes the quality-detectability frontier of GaussMark close to that of inference-time watermarking, remains robust to paraphrasing and fine-tuning attacks, and exhibits strong generalization: a model fine-tuned on one dataset retains substantial watermark detection power on unseen datasets. Together, these results establish MarkTune as a general strategy for embedding robust, high-quality watermarks into open-weight LMs.

Large Language Models (LLMs) are trained to refuse to respond to harmful content. However, systematic analyses of whether this behavior is truly a reflection of its safety policies or an indication of political censorship, that is practiced globally by countries, is lacking. Differentiating between safety influenced refusals or politically motivated censorship is hard and unclear. For this purpose we introduce PSP, a dataset built specifically to probe the refusal behaviors in LLMs from an explicitly political context. PSP is built by formatting existing censored content from two data sources, openly available on the internet: sensitive prompts in China generalized to multiple countries, and tweets that have been censored in various countries. We study: 1) impact of political sensitivity in seven LLMs through data-driven (making PSP implicit) and representation-level approaches (erasing the concept of politics); and, 2) vulnerability of models on PSP through prompt injection attacks (PIAs). Associating censorship with refusals on content with masked implicit intent, we find that most LLMs perform some form of censorship. We conclude with summarizing major attributes that can cause a shift in refusal distributions across models and contexts of different countries.

Unified Multimodal Models (UMMs) have shown impressive performance in both understanding and generation with a single architecture. However, UMMs still exhibit a fundamental inconsistency: understanding favors compact embeddings, whereas generation favors reconstruction-rich representations. This structural trade-off produces misaligned decision boundaries, degraded cross-modal coherence, and heightened vulnerability under distributional and adversarial shifts. In this paper, we present UniGame, a self-adversarial post-training framework that directly targets the inconsistencies. By applying a lightweight perturber at the shared token interface, UniGame enables the generation branch to actively seek and challenge fragile understanding, turning the model itself into its own adversary. Experiments demonstrate that UniGame significantly improves the consistency (+4.6%). Moreover, it also achieves substantial improvements in understanding (+3.6%), generation (+0.02), out-of-distribution and adversarial robustness (+4.8% and +6.2% on NaturalBench and AdVQA). The framework is architecture-agnostic, introduces less than 1% additional parameters, and is complementary to existing post-training methods. These results position adversarial self-play as a general and effective principle for enhancing the coherence, stability, and unified competence of future multimodal foundation models. The official code is available at: https://github.com/AIFrontierLab/UniGame