Research Papers

LLM supervision systems, namely input/output moderation filters and jailbreak detectors, are the primary safeguard against misuse in deployed AI applications, yet existing benchmarks are often vendor-biased, omit cost and latency, and rarely compare specialized guardrails against repurposed generalist LLMs. We present BELLS-O (Benchmark for the Evaluation of LLM Supervision Systems, Operational), the first independent operational benchmark of LLM supervision systems. BELLS-O evaluates 28 systems from 17 providers: every major specialized guardrail (e.g., LlamaGuard-4, ShieldGemma-2, Lakera Guard) and frontier generalists repurposed as supervisors (e.g., GPT-5.4, Claude Sonnet 4.6, Grok-4.1), jointly on detection rate, false-positive rate, latency, and monetary cost. We cover input/output moderation across 11 harm categories and jailbreak detection across 13 attack techniques, using in-house datasets built from handcrafted prompts, expert-curated samples, and quality-controlled synthetic generation. To suppress latent generator fingerprints in synthetic data, every generated sample is paraphrased. Mapping the Pareto frontier reveals use-case-dependent tradeoffs. On content moderation, specialized supervisors are operationally dominant: top systems match frontier LLMs on detection (~95% vs. 94%) at comparably low false-positive rates (<=2%), while running 5-10x faster and ~10x cheaper. On jailbreak detection, the tradeoff shifts: frontier LLMs achieve higher detection and lower false-positive rates but at 10-50x higher cost and 5-10x higher latency. We release the benchmark, framework, leaderboard, and datasets as the first vendor-neutral basis for selecting safeguards under real deployment constraints.

Modern artificial intelligence (AI) systems present profound new risks to child safety. AI is increasingly being misused to create AI-generated child sexual abuse material, facilitate child sexual exploitation, and reduce barriers to harm. In this paper, we argue that protecting children from AI-facilitated sexual abuse requires new approaches to AI safety. Existing safety techniques assume data accessibility, transparency, and evaluation practices that are incompatible with the ethical and legal constraints surrounding child sexual abuse material. We examine how these constraints create new technical challenges, such as limitations on dataset auditing, red teaming, and fine-tuning prevention. In turn, we outline *15 open problems* in online child sexual exploitation and abuse across the AI development lifecycle, from dataset curation and model design to deployment and long-term maintenance. We propose targeted recommendations for researchers, developers, and policymakers to bridge the gap between theoretical AI safety and the realities of child protection. Our work aims to reframe preventing AI-facilitated child sexual abuse as a central, safety-critical dimension for AI research, motivating work that translates responsible AI principles into concrete safeguards against the exploitation of children.

In sixth-generation (6G) networks, billions of cyber-physical systems (CPSs) - autonomous vehicles, smart grids, industrial robots, and remote-surgical equipment - will run over ultra-reliable low-latency slices, collapsing the gap between a remote breach and physical harm to milliseconds, a budget perimeter firewalls and centralised security operations centres cannot meet. This survey reframes 6G CPS security as a closed-loop, AI-native pipeline that senses at the multi-access edge computing (MEC) tier, using minute-scale call-detail records (CDRs) for baseline learning and sub-millisecond RAN/Open-RAN (O-RAN) telemetry for the latency-critical path. It decides locally with compressed deep models, mitigates network-wide via SDN, NFV, and O-RAN controllers, and retrains through federated learning (FL) and digital-twin (DT) replay. We formalise a per-slice, tail-bounded latency contract on the sense, detect, and mitigate stages, enforced at a slice-dependent tail percentile (p99 for safety-critical URLLC slices). Organising 128 peer-reviewed studies (2017-2026) under a PRISMA 2020 protocol, we (i) map the 6G/CPS threat surface to MITRE ATT&CK and a CDR-observable feature space; (ii) unify edge anomaly detection and DDoS classification across twelve datasets and statistical, graph, and transformer models; (iii) synthesise SDN/NFV/O-RAN primitives into one closed-loop reference architecture; (iv) treat FL, large language models (LLMs), DT, post-quantum cryptography (PQC), zero-trust architecture (ZTA), and explainable AI as cross-cutting enablers, not parallel pillars; and (v) consolidate open problems into five directions spanning data, latency, trust, standardisation, and evaluation.

High-stakes enterprise document generation, including financial dispute narratives, compliance notices, and audit summaries, demands schema correctness, policy compliance, and low-latency operation at scale. Prior to a unified guardrail layer, production systems often stitched together separate PII redaction, content moderation, and format validation steps, leading to fragmented logic, slower request paths, and higher operational cost. We present a guardrail orchestration layer for text and image inputs that couples multi-candidate generation with an explicit compliance score used for early exit. The framework runs configurable parallel generation heads, scores candidates against weighted guardrails including PII detection, content moderation, schema constraints, and domain rules, and returns the best-scoring output with selection metadata. The available operational readout reports 5 attempts within 20 seconds and 91 percent compliance. For payments dispute defense summaries, we analyze aggregate operational scenario readouts rather than a randomized A/B test. Variable cohorts show higher count win rates than controls overall, 301/659 versus 536/1548, corresponding to +11.0 percentage points with 95 percent confidence interval [6.6, 15.5] and p < 0.001, and for adjusted item-not-received cases, +7.5 percentage points with 95 percent confidence interval [0.2, 15.7] and p = 0.045. Fraud and local evidence-ranking deltas are directionally positive but not statistically significant from the aggregate count data. We also report reviewer-calibrated Responsible-AI evidence-quality signals from 770 generated-evidence reviews and a 70-case OCR slice, and document the reproducibility boundary through the request interface, scoring logic, pseudocode, and operational evidence boundary.

Large language model (LLM) unlearning has emerged as a crucial post-hoc mechanism for privacy protection and AI safety, yet auditing whether target knowledge is truly erased remains challenging. Existing output-level metrics fail to detect when this knowledge remains recoverable from internal representations. Recent white-box studies reveal such residual knowledge but often rely on auxiliary training or dataset-specific adaptations, leaving no generalizable metric. We close this gap with the Unlearning Depth Score (UDS), a metric that quantifies the mechanistic depth of unlearning via activation patching. UDS first identifies layers that encode the target knowledge using a retain model baseline, then measures how much of it is erased in the unlearned model on a 0-1 scale. In a meta-evaluation across 20 metrics on 150 unlearned models spanning 8 methods, UDS achieves the highest faithfulness and robustness, confirming our causal approach as the most reliable for unlearning evaluation. Case studies further show that UDS uncovers residual knowledge obscured from observational metrics by representational shifts, with erasure depth varying across prompt types. We provide guidelines for integrating UDS into existing benchmarking frameworks and streamlining the evaluation pipeline. Code and data are available at https://github.com/gnueaj/unlearning-depth-score.

Affordances and permissions are promising and timely safety levers for mitigating Loss of Control (LoC) threats in high-stakes deployment contexts, such as national security. Deployers in defense and intelligence could rely on several approaches to identify which affordances and permissions should be prioritized, such as structured threat modelling, pre-deployment agentic evaluations, post-deployment continuous monitoring, and AI safety cases. This paper proposes a complementary and empirical methodology that leverages existing use-case-specific benchmarks: backchaining LoC mitigations from the errors an AI system makes on national security benchmarks. The approach proceeds in three steps and allows national security deployers to start building LoC mitigations today, from evidence they can generate themselves. First, deployers evaluate AI systems on mission-specific benchmarks approximating real use-cases. Second, deployers concentrate on the incorrect responses that the AI system provides to the benchmark questions, and backchain the affordances and permissions that would enable the AI system to cause downstream harm if it pursued the actions described in the incorrect answers. Third, deployers intervene selectively on those affordances and permissions, bottlenecking the paths to harm while preserving the AI system's ability to carry out the correct action. We illustrate this methodology through a demonstrative benchmark question on derivative security classification.

Explainable Machine Learning for Phishing Detection on Heterogeneous Datasets with MCP-Enabled Deployment

Nikhil Kumar Dora, Sumit Kumar Tetarave, Rishikesh Sahay +2

With the growth in digital transformation and Internet usage, the Social Engineering techniques such as Phishing have become a major concern for the users and the organizations. Phishing attacks involve deceptive techniques to trick users into revealing confidential information that causes financial loss and reputation damage to organizations. According to report of Verizon, 36% of all data breaches involved phishing, highlighting the need for intelligent, adaptive, and explainable security mechanisms. This paper examines the efficiency of different machine learning algorithms in phishing detection on heterogeneous phishing datasets that include a publicly available UCI dataset, our generated datasets using tools such as EvilGinx and Zphisher, and AI generated datasets. Moreover, this work incorporates explainable AI (XAI) techniques such as Information Gain, SHAP (SHapley Additive Explanations), and LIME (Local Interpretable Model-Agnostic Explanations) to examine the most influential features impacting classification outcomes. To support practical deployment, this work also incorporates an MCP-based phishing URL detection system that offers real-time URL analysis, feature extraction, confidence-based classification, and AI-assisted security interpretation. The experimental results demonstrate that among classical models the highest accuracy is obtained by Logistic Regression at 92.44%, among ensemble models CatBoost achieved the highest accuracy at 95.01%, among neural network CNN achieved an accuracy of 94.02%, and among transformer-based models, DistilBERT got the highest accuracy at 99.78%

Jailbreaking attacks on large language models pose a significant threat to AI safety by enabling the generation of harmful or restricted content. While prior work has explored both handcrafted and automated jailbreak strategies, the potential for compositional interaction between simple attacks remains underexplored. This paper presents a systematic study of mutator chaining, in which weak jailbreak transformations are applied sequentially to characterize how they interact: whether they reinforce one another, interfere destructively, or produce no meaningful change. We implement twelve baseline mutators and evaluate all ordered pairs on a benchmark of harmful prompts against three popular LLM models. Our framework introduces metrics for completeness and validity that capture both transformation persistence and attack effectiveness. Results reveal that the interaction landscape is highly non-uniform, while most combinations fail to outperform individual mutators, exhibiting destructive interference or structural incompatibility, a small fraction produce synergistic effects that improve attack success rates. Equally important, the prevalent failure modes reveal structural properties of safety alignment that are not apparent from single-strategy evaluations. These findings highlight the nuanced dynamics of adversarial prompt composition and offer new insights for building more robust safety defenses.

The Attacker in the Mirror: Breaking Self-Consistency in Safety via Anchored Bipolicy Self-Play

Gabriele La Malfa, Emanuele La Malfa, Saar Cohen +4

Self-play red team is an established approach to improving AI safety in which different instances of the same model play attacker and defender roles in a zero-sum game, i.e., where the attacker tries to jailbreak the defender; if self-play converges to a Nash equilibrium, the model is guaranteed to respond safely within the settings of the game. Although the parameter sharing enforced by the use of the same model for the two roles improves stability and performance, it introduces fundamental theoretical and architectural limitations. We show that the set of Nash equilibria that can be reached corresponds to a broad class of behaviours that includes trivial always refuse strategies and oracle-like defenders, thus limiting practical applicability. We then show that when attacker and defender share and update the same base model, the dynamics collapse to self-consistency, so that attacks do not enforce adversarial pressure on the defender. In response, we propose Anchored Bipolicy Self-Play, which trains distinct role-specific LoRA adapters on top of a frozen base model, thereby maintaining stable optimisation while preserving adversarial pressure through explicit role separation. In relation to standard self-play, we show up to 100x greater parameter efficiency than finetuning and consistent improvements in safety compared to self-play fine-tuned models. We evaluate on Qwen2.5-{3B, 7B,14B}-IT models across widely used safety benchmarks, showing improved robustness without loss of reasoning ability. Cross-play experiments further show that our attacker and defender models are superior to self-play in terms of adversarial defence and safety.

Current adversarial robustness methods for large language models require extensive datasets of harmful prompts (thousands to hundreds of thousands of examples), yet remain vulnerable to novel attack vectors and distributional shifts. We propose Latent Personality Alignment (LPA), a sample-efficient defense that achieves robustness by training models on abstract personality traits rather than specific harmful behaviors. Using fewer than 100 trait statements and latent adversarial training, LPA achieves comparable attack success rates to methods trained on 150k+ examples, while maintaining superior utility. Critically, LPA generalizes better to unseen attack distributions, reducing misclassification rates by 2.6x compared to baseline across six harm benchmarks -- without ever seeing harmful examples during training. Our results demonstrate that personality-based alignment offers a principled approach to building robust defenses with minimal cost.

The security of open-source software repositories is increasingly threatened by next-gen software supply chain attacks. These attacks include multiphase malware execution, remote access activation, and dynamic payload generation. Traditional Machine Learning (ML) detectors struggle to detect these attacks due to the high-dimensional and sparse nature of dynamic behavioral data, including system calls, network traffic, directory access patterns, and dependency logs. As a result, these data characteristics degrade the performance, stability, and explainability of ML models. These challenges have made Deep Learning (DL) a promising alternative, given its success across various domains and its potential for modeling complex patterns. This paper presents eDySec, a DL-based efficient, stable, and explainable framework for dynamic behavioral analysis to detect malicious packages. Using the QUT-DV25 dataset, which captures both install-time and post-installation behaviors of packages, we evaluate DL models and investigate feature sets to identify the most discriminative attributes for enabling efficient malicious package detection. Additionally, model stability analysis and explainable AI techniques are incorporated into the detection pipeline to enable stable, and transparent interpretations of model decisions. Experimental results demonstrate that eDySec significantly outperforms the state-of-the-art frameworks. Specifically, it halves feature dimensionality while lowering false positives by 82% and false negatives by 79%. It also improves accuracy by 3%, achieves near-perfect stability, and maintains an inference latency of 170ms per package. Further analysis reveals that feature and model selection play a critical role, as certain combinations degrade performance. Ultimately, this study advances the understanding of the strengths and limitations of dynamic analysis against next-gen attacks.

The rapid integration of artificial intelligence (AI) into Internet of Things (IoT) and edge computing systems has intensified the need for robust, hardware-rooted trust mechanisms capable of ensuring device authenticity and AI model integrity under strict resource and security constraints. This survey reviews and synthesizes existing literature on hardware-rooted trust mechanisms for AI-enabled IoT systems. It systematically examines and compares representative trust anchor mechanisms, including Trusted Platform Module (TPM)-based measurement and attestation, silicon and FPGA-based Physical Unclonable Functions (PUFs), hybrid container-aware hardware roots of trust, and software-only security approaches. The analysis highlights how hardware-rooted solutions generally provide stronger protection against physical tampering and device cloning compared to software-only approaches, particularly in adversarial and physically exposed environments, while hybrid designs extend hardware trust into runtime and containerized environments commonly used in modern edge deployments. By evaluating trade-offs among security strength, scalability, cost, and deployment complexity, the study shows that PUF-based and hybrid trust anchors offer a promising balance for large-scale, AI-enabled IoT systems, whereas software-only trust mechanisms remain insufficient in adversarial and physically exposed settings. The presented comparison aims to clarify current design challenges and guide future development of trustworthy AI-enabled IoT platforms.