Papers

Deep neural networks (DNNs) are used in a variety of real-world applications including, for example, image classification and speech recognition. The inference accuracy of DNN implemented on hardware in integrated circuits (ICs) degrades under phenomena such as transistor aging. Aging slows down the switching speed of transistors, resulting in system-level timing violations due to unsustainable clocks. To maintain reliability for the entire projected lifetime, designers add guardbands to prevent timing violations; however, adding large timing guardbands causes losses in performance (speed or throughput). This chapter provides a detailed discussion of the effects of long-term and short-term transistor aging on DNN inference accuracy. Furthermore, to mitigate aging effects on DNN's accuracy and keep them at bay, a methodology for aging-aware retraining is presented in order to generate a resilient DNN even when aggressive (i.e., smaller than required) guardbands are used. This improves the inference accuracy of the DNNs even in the presence of aging-induced degradation. These effects are discussed in this chapter along with mitigation strategies on a hardware implementation of a DNN for image classification on an off-the-shelf image dataset. The application of short-term aging as an excitation mechanism for the detection of hardware Trojans in integrated circuits is also briefly discussed.

Current AI agent observability is structurally compromised: the entity producing the activity log is the same entity whose activity is being logged. A compromised or buggy agent can omit, alter, or fabricate its own traces, and the operator running the agent has no independent way to detect tampering. We propose a class of protocols that resolves this by inverting the trust boundary: the service that receives an agent's call signs a receipt of what it observed using its own key, encrypts the receipt to the agent's owner, and publishes it to a public transparency log. The owner reconstructs a tamper-evident trail without trusting the agent or its operator. We instantiate the class as Sello, a protocol combining four properties absent in any current system: (P1) receiver-side signing, (P2) HPKE encryption to an owner public key bound to the authorization token via JWS, (P3) publication to a witness-cosigned Merkle log, and (P4) owner-side discovery by token reference. We describe the protocol, analyze its security under an adversary that controls the agent and its operator, present microbenchmarks of the cryptographic operations, and situate Sello among adjacent receipt-protocol work (Signet, AgentROA, Agent Passport System, draft-farley-acta, SCITT). We discuss known limitations including the suppression attack, service collusion, and the adoption-incentive problem.

File-type classification underlies many workflows like malware triage, forensic carving, packet inspection, and storage indexing. Learned systems such as Google's Magika assume whole-file access at a known offset, so they break on the inputs many of these tasks actually produce, like a single packet payload, a header-less carved fragment, a random disk block, or a chunked upload. We introduce MimeLens, a family of small BERT-style encoders pretrained on binary content from windows sampled at a uniformly random offset within each file, with no privileged head-of-file position, in standard- and short-context variants. A byte chunk goes in from anywhere in a file, no header needed and no fixed size; out comes one of libmagic's 125 MIME labels. On the clean head of complete files, MimeLens beats Magika v1.1 by +10.7 pp top-1 on libmagic-labeled data, and it keeps classifying where Magika cannot: from a single mid-stream UDP packet, and more than twice as accurately as libmagic and Magika on random mid-file disk blocks. The cost is latency: MimeLens runs roughly one to two orders of magnitude slower per sample on CPU than Magika, though it matches on consumer GPUs or in batch. All trained checkpoints are released on Hugging Face (mjbommar/mimelens-001-*).

Safety alignment in large language models (LLMs) is fragile in part because it is often shallow: fine-tuning mainly reshapes the model's behavior near the first few output tokens. We argue that this phenomenon can be understood through autoregressive consistency, the tendency of next-token prediction to preserve and extend the current response trajectory consistently. By analyzing the learning dynamics of safety alignment, we show that autoregressive consistency can concentrate alignment updates on early tokens, offering a mechanistic explanation for shallow safety alignment. The same mechanism also predicts a broader class of attacks on LLMs: attacks that induce harmful continuation states at arbitrary positions in the output trajectory. As a concrete example, we introduce random insertion attack, which inserts a short harmful span into an otherwise safe refusal trajectory and exploits autoregressive consistency to sustain the resulting harmful branch, thereby bypassing safety alignment. Notably, a short harmful span can redirect the generation to be harmful even after a long refusal prefix, highlighting autoregressive consistency as a potential broader failure mechanism. This suggests that safety alignment should also break harmful autoregressive consistency throughout the output trajectory. We therefore propose adversarial safety alignment, an initial framework based on worst-case harmful continuation states, and instantiate it with random worst-insertion training. Overall, our results suggest that autoregressive consistency should be treated as a central consideration in both safety alignment and attack design.

LLM agents often place sensitive credentials in the same context window as untrusted retrieved content, creating a direct path for indirect prompt injection to induce credential exfiltration. We study this failure mode through three complementary defenses. First, we ask whether activation probes can detect credential access before output tokens are emitted. Second, we construct honeytokens from format-specific character models and calibrate detection with split conformal prediction. Third, we treat multi-turn exfiltration as a cumulative information-flow problem and track an estimated leakage budget across conversation turns. In controlled experiments on open-weight models, activation features separate benign and credential-seeking prompts with high accuracy, including under held-out encoding transformations. In a small synthetic multi-turn suite, cumulative accounting detects attacks that per-turn detectors miss. These results are preliminary: the multi-turn benchmark is in-house and small, the activation method requires white-box access, and the information estimator provides a practical signal rather than a formal upper bound. Still, the results suggest that credential-exfiltration defenses should combine pre-output monitoring, calibrated canary detection, and temporal leakage accounting rather than relying only on text-level output filters.

Agent systems execute through runtimes with very different control points: local coding tools, framework SDKs, managed agent platforms, API gateways, and observer-only integrations. A high-risk action such as publishing data externally may therefore appear as a shell command in one runtime, a tool call in another, and a hosted session transition in a third. This makes it difficult to answer a basic governance question consistently: what action was authorized, under whose authority, with what approval semantics, and with what evidence after execution? This paper presents Proof-Carrying Agent Actions (PCAA), a runtime-neutral governance model centered on an action certificate rather than on a vendor-native session record. PCAA organizes control around five checkpoints: pre-action admissibility, action open, assumption capture, approval, and outcome closure. It binds these checkpoints to a portable action envelope, runtime and approval receipts, and replay-ready proof. The model is extended in two practical ways: the certificate is externality-aware, carrying boundary facts such as destination visibility and account provenance, and approval is described by explicit enforceability classes rather than by a single reviewed or unreviewed bit. We study the model through a reference implementation in a heterogeneous agent control plane and a disclosure-bounded evaluation protocol. On a protected benchmark expanded from 24 executable seeds to 96 traces across four runtime families, PCAA preserves route quality while exposing distinct failure modes under ablation. The paper contributes a systems formulation of runtime governance around certificate-bearing actions and an implementation-grounded account of how that formulation can remain portable under runtime churn without collapsing into vendor-specific control surfaces.

Reinforcement learning (RL) has become a dominant post-training paradigm, enabling large language models (LLMs) to learn from rewards. We observe that societal regulations are structurally similar to reward functions. They define measurable outcomes, thresholds, and exceptions, while often leaving institutional intent only partially specified. We hypothesise that the RL training process may exploit these gaps and therefore ask whether models' well-known tendency to hack reward functions during RL can scale into a more consequential failure mode named societal hacking: discovering loopholes in the rules society runs on. To study this phenomenon, we introduce SocioHack, a sandbox of 72 societal environments, and find that within these environments, reward hacking naturally emerges and leads to regulatory loophole discovery. Models learn to hack the social rules and generate strategies that remain technically compliant while defeating regulatory intent, and current LLM safeguards provide only limited mitigation. Therefore, collecting in-the-wild feedback for model training requires greater caution, and we need a next-generation post-training paradigm for safely iterating LLMs in real society.=

As language models increasingly consume one another's outputs, covert influence – a phenomenon where a sender's payload (the behavioral disposition it is conditioned to propagate) transfers to a receiver through carriers undetectable by humans – becomes a growing risk. We characterize this risk across three interfaces: supervised fine-tuning, on-policy distillation, and in-context learning, and find that they vary in the scale of influence achievable without leaving behind human-visible traces. Using inference-time per-sample attribution scores, we study covert influence across all three interfaces with the ability to select carriers that amplify training-time influence, unlocking payload transfers that prior work could not achieve. We further provide evidence that covert influence with natural-language carriers is a distinct phenomenon from prior studies using number carriers, as the latter is more resistant to human detection and less portable across model families. Together, these results suggest that the risk surface for covert influence is broader than previously recognized, and we study pointwise attribution scoring methods as a tool to investigate and mitigate it.

Existing privacy analyses for Graph Neural Networks (GNNs) largely inherit assumptions from non-graph settings, overlooking structural correlations and stochastic training-graph sampling. In particular, node-dependent priors make type-I and type-II errors alone insufficient to characterize the best membership inference test. To address this, we introduce Bayesian Membership Privacy (BMP), a sampling-aware formulation of node-level membership privacy that incorporates node-dependent priors and treats graph sampling probabilities as part of the adversary's knowledge. BMP casts membership inference as a Bayesian hypothesis test and accordingly quantifies membership privacy in terms of posterior membership probability. We explore theoretical properties of BMP in relation to the existing definitions in the literature. We further propose a practical, sampling-aware auditing mechanism to estimate the parameters of BMP as a measure of node-level privacy leakage in GNNs. We conduct experiments on benchmark graph datasets and show that BMP yields fine-grained privacy insights that are not visible through global attack accuracy alone.

As LLMs become increasingly woven into everyday workflows, user queries sent to cloud hosted LLMs routinely mix task-essential content with task non-essential sensitive disclosures, yet type based PII redaction is context agnostic and may raise two issues: over disclosing untyped sensitive context and over removing answer bearing spans. We recast privacy preserving query rewriting under Contextual Integrity: a span should be forwarded only if it is necessary for the task. We introduce DelegateCI-Bench, the first task based Contextual Integrity benchmark for privacy-conscious delegation, comprising 3,167 samples that combine high quality synthetic data spanning 11 tasks and 20 task types, WildChat based real user queries, and a medical challenge set with dense sensitive information. Building on this benchmark, we propose a CI-guided reinforcement learning framework that converts essential and non-essential sensitive spans into verifiable optimization signals, and train a query rewriter to preserve task critical information while suppressing unnecessary sensitive disclosure. Experiments show that our learned rewriter achieves the best privacy-utility tradeoff, achieving up to +10.1 average utility over on-device baselines.

The evolution of LLMs into tool-enabled agents creates a new class of safety challenges associated with real-world execution rather than simple text generation. Existing alignment methods often rely on coarse refusal signals or static supervision, making it difficult to balance safety with useful tool execution across diverse agentic risks. We introduce RUBAS, a rubric-based reinforcement learning framework for agent safety. RUBAS decomposes agent behavior into four dimensions: tool-use safety, argument safety, response safety, and helpfulness. These structured rubrics provide fine-grained and interpretable rewards over complete agent trajectories, enabling reinforcement learning to optimize safe tool use while preserving task completion. Extensive experiments across multiple agent safety benchmarks and models show that RUBAS improves safety over standard alignment baselines, reduces tool-grounded hallucinations, and maintains competitive utility. Our results suggest that multi-dimensional rubric rewards provide an effective training signal for aligning LLM agents in safety-critical tool-use settings.

Large language model (LLM) agents can persist across tasks, acquire memory, activate Skills, synthesize tools, fork child processes, attach remote resources, and commit checkpoints as reusable images. These mechanisms expand the action surface after deployment and create authority-escalation and data-exfiltration risks when visibility is mistaken for permission. We present Agent libOS, an agent-native library OS substrate that separates three planes. Operation admission combines process identity, Task Authority ceilings, typed Capabilities, policy or Human approval, budgets, and concrete primitives. Information-flow admission propagates labels and immutable source references, resolves Host-registered Sinks, and requires an exact one-shot Human release for conditional high-sensitivity egress. Durable causal evidence records intent, outcomes, accounting, and causal links but never grants authority. Thus, the model-visible action surface may evolve without implicitly expanding resource authority or permitted information flows. The implementation provides persistent processes, Object Memory, Skills, syscall-mediated JIT Tools, images and checkpoints, typed providers, Human queues, budgets, and durable recovery. Provider-backed effects use a prepare-dispatch-settle protocol that exposes ambiguity and prevents blind replay. In source-bound evaluation, 33/33 deterministic full-runtime tasks pass both task and safety oracles. Across 12 canonical real-model runs, observed safety and strict utility are 12/12. In a paired 30-run Skill projection study, the observable-state oracle passes in all runs, with 13/15 fully correct runs in each arm. These results describe the evaluated model/provider configuration. Agent libOS does not prevent prompt injection, provide kernel-grade sandboxing, or roll back irreversible external effects.

A computer worm is malware that spreads on a network by replicating itself from one machine to another. Traditional worms, like WannaCry, exploited predetermined vulnerabilities, and their spread can be halted by patching those vulnerabilities. Here we show that artificial intelligence (AI) agents enable a fundamentally new threat: a worm that generates tailored attack strategies to each target it encounters. The worm parasitically uses compromised machines to run open-weight large language models (LLMs) to sustain its reasoning, or extend its reach for further attacks. Deployed on a network of machines spanning Linux, Windows, and IoT (Internet of Things) devices, the worm propagated by exploiting common, real-world corporate network vulnerabilities. Since the worm is powered by stolen compute, the attacker's marginal cost per new infection is zero. This creates a destabilizing economic asymmetry between attackers and defenders. Moreover, because the worm requires no commercial AI platform, centralized safety controls, such as service refusals or rate limiting, are structurally irrelevant. Our results demonstrate that self-sustaining AI-driven cyber-threats are no longer theoretical. We must prepare for autonomous generative adversaries: malware systems that propagate without human operators and are defined not by fixed exploit code, but by the capacity to reason about targets, adapt to observations, and synthesize attack logic in real time.

We propose PURGE, a machine unlearning algorithm built on a simple but an under-exploited observation: continual learning (CL) and machine unlearning (MU) which are fundamentally dual problems. CL tries to learn new tasks without forgetting old ones; MU tries to erase specific data without hurting retained performance representing the same underlying tension in opposite directions. PURGE leverages this duality by adapting gradient projection from A-GEM (Chaudhry et al., 2019) so that every unlearning step is constrained to not increase the retain-set loss. On top of this, it performs multi-layer representation erasure, pushing forget-set activations in intermediate layers towards the retain distribution to remove information from hidden representations rather than just suppressing it at the output. A key design choice is the retain-confusion target: rather than pushing forget outputs toward the uniform distribution, which we found to be surprisingly easy for membership inference attacks to detect, we instead target the model's natural confusion pattern on retain data. This makes the unlearned model hard to distinguish from one retrained from scratch. Two self-regulating stopping criteria (a retain-loss budget and a forget-accuracy target) let the algorithm decide on its own when to stop, removing the need for manual epoch tuning. In experiments on five datasets (CIFAR-10, MNIST, SVHN, STL10, PathMNIST) across 22 class-level forgetting tasks, PURGE consistently keeps retain accuracy above 96% while achieving MIA AUROC close to 0.5 (the ideal), outperforming gradient ascent, KL-uniform, and several published baselines on the privacy-utility frontier.

We initiate the study of the algorithmic complexity of finding collisions in single-layer binary neural networks. Given a random matrix $\mathbf{A} \in \mathbb{R}^{m\times n}$, an input $\mathbf{x} \in \{-1,1\}^n$ is mapped to a binary output vector $\varphi(\mathbf{A}\mathbf{x})\in \{-1,1\}^m$, where $\varphi$ is an activation function with constant behavior on $[\kappa, \infty)$ for some threshold $\kappa \geq 0$. We identify the threshold scale $\kappa=\Theta(1/\sqrt{\alpha})$, where $\alpha=m/n$, as separating two complementary phenomena. When $\kappa \ll 1/\sqrt{\alpha}$, we give a simple online algorithm that efficiently produces extensive collisions. When $\kappa \gg 1/\sqrt{\alpha}$, for a natural randomized non-periodic activation and suitable oscillation complexity, we prove that the extensive-collision space exhibits an overlap gap property (OGP), yielding an exponential lower bound against online algorithms. Ours is the first work to use the overlap gap property as a rigorous criterion for collision resistance. The key difference between collision finding and average-case search is that collision finding has a new “worst-case” aspect: the collision finder has full control over the choice of colliding pairs. Our lower bound is proved in the online model; extending such guarantees to broader classes of algorithms, including spectral, algebraic, lattice-based, or quantum methods, remains an open direction.

AI losses that arise through an insured organization's generative or agentic AI system require state reconstruction, not merely event reconstruction, because the relevant state changes as the system reasons, retrieves, calls tools, and acts. The relevant question is not only what loss occurred, but what the system was allowed to do, what it actually did, and whether that reconstructed loss can support insurance claim recovery. This paper addresses losses in which the insured's AI system is in the causal chain, including externally triggered failures such as prompt injection, retrieval-augmented generation (RAG) poisoning, malicious tool output, credential misuse, and data poisoning. Specifically, this paper introduces CER, a use-case-level diagnostic for AI residual risk transfer. C (control boundary) asks whether the system had an enforceable operating envelope. E (evidence reconstruction) asks whether the system state and causal chain can be reconstructed from retained artifacts. R (insurance response) asks whether the reconstructed loss is insured: whether insurance coverage is available in the market and placed for the insured, together with the proof needed to support insurance claim recovery. The paper makes three contributions: it defines the AI-specific reconstruction problem, operationalizes that problem through CER, and specifies claim-grade evidence for AI reconstruction. Public examples include the reported PocketOS and Replit agentic database-deletion incidents and Moffatt v. Air Canada as an adjudicated output/reliance case. Keywords: AI systems; CER framework; residual risk transfer; agentic AI; generative AI; AI insurance; evidence reconstruction.

Decentralized verifiable credential systems have seen limited deployment in practice. Existing constructions, built on zero-knowledge proofs, are complex, application-specific, and largely restricted to predicates over structured data. We present Privately Inferred Credentials ($\pi$Creds): privacy-preserving, legacy-compatible, decentralized verifiable credentials generated by trusted LLM inference over authenticated data. LLMs' ability to semantically reason over unstructured data substantially expands the range of claims $\pi$Creds can certify over existing credential systems. The use of LLMs also introduces new application-level threats, which we formalize through two problems: the Source-Constrained Adversarial Example (SCAE) problem, which captures robustness against adversaries that manipulate authenticated data to obtain misleading credentials, and the Authenticated Covert Predicate Poisoning (ACPP) problem, which captures privacy leakage through adversarial model selection. We characterize applications of $\pi$Creds over user data, and a novel class of credentials over proprietary software that certifies properties of a service without revealing its source code. Our prototype supports issuing credentials over live financial, health, email, and code sources, and we empirically study the SCAE and ACPP threats on a product expertise credential over real financial data.

Vision-language-action (VLA) policies are often treated as checkpoint-defined objects: if the weights, prompt, and benchmark suite match, the deployment is assumed to be the same policy. Robot execution breaks this assumption because the same normalized model output can become a different physical action after action unnormalization and controller conventions are applied. This creates a deployment-safety gap: safety review can certify the checkpoint while missing the executable robot policy that reaches the controller. We formalize this gap as an executable policy specification problem: a VLA policy includes the learned model, action representation, metadata-selected unnormalizer, and controller-facing conventions. Under this view, identical checkpoints can be executable-inequivalent. For quantile-style action normalization, we derive a closed-form metadata mismatch transform and an ExecSpec certificate that measures action-space semantic drift without model inference or rollout. On LIBERO-Goal replay, substituting a plausible sibling metadata key yields mean drift 0.199 over six non-gripper action dimensions and reduces success from 28/28 to 2/28 under full substitution. On LIBERO-Spatial replay, the same substituted key reduces success from 26/26 to 0/26. The same full-substitution protocol gives 0/28 success for all four Object substitutions and 0/23 or 1/23 success on Long. Identity-key, replay-validity, no-op filtering, raw-vs-correct replay, mask/gripper, synthetic upper-bound, and OpenVLA-style unnormalizer interface checks rule out several simpler explanations. These results do not certify closed-loop or hardware safety. They support a narrower deployment-safety view: action-space metadata is part of the executable policy and should be checked before rollout.

Android malware detection increasingly relies on collecting and processing sensitive user data, including device identifiers, network artifacts, and runtime traces, while privacy is too often treated as a secondary concern. Existing privacy-aware approaches typically enforce privacy after data collection, for example, through anonymization, encryption, or federated learning, yet still require access to user information and therefore demand a high level of user trust in systems that already operate with privileged access to device activity. We argue that this requirement should be removed rather than managed. Android malware detection should be privacy-aware by design, so that effective analysis does not depend on sensitive data being accessed in the first place. To this end, we first formalize a set of design requirements for privacy-by-design detection and then implement each requirement in a comprehensive pipeline. First, static analysis is performed to extract relevant data from each APK, following the Drebin representation, which is then submitted to an SVM after vectorization. The model is equipped with a dual-reject threshold rule that either commits to a confident decision or defers uncertain samples to a dynamic analysis stage within a sandboxed environment, so that genuine user information never enters the analysis loop. Results confirm that, on a temporally split dataset spanning from 2024 to 2025, the pipeline achieves an F1 score of 0.87 with the first static analysis stage, deferring only 6.7% of test samples to secondary dynamic analysis. Additionally, dynamic sandboxing helps recognize applications' maliciousness with high confidence without extracting any sensitive data. These results demonstrate that strong detection performance is achievable without sacrificing user privacy.

A publisher who releases check-in trajectories inadvertently publishes a strong predictor of every user's future locations. We address this risk by generating unlearnable trajectories, perturbed sequences that yield victim models with degraded next-Point-of-Interest (next-POI) accuracy on clean test inputs. Direct ports of image-domain unlearnable examples fail on two counts. The published data must remain geographically and semantically plausible, and the perturbation must resist purification adversaries that exploit the structure of randomized defences. We propose Ghost, a manifold-aligned framework whose perturbations look like plausible human check-in sequences yet leave no learnable signal behind. Ghost steers each substitution onto the real-trajectory manifold through a frozen trajectory language model, so a denoising-bridge adversary has nothing to invert and a context-free frequency-table adversary recovers a near-uniform distribution. Across two standard benchmarks, and four attacker postures, Ghost achieves protection-gap competitive with the strongest deterministic baseline (PGD) while attaining the lowest restored accuracy under the bigram adaptive purification adversary on both datasets, and lies within one per-cell standard deviation of PGD on the protection-versus-purification-resistance plane. Ablations confirm the manifold prior subsumes the entropy-floor knob of prior randomized defences, with the frequency-table adversary's survival gap remaining within 0.04 even when twenty percent of the pairs are leaked.

Integrated Circuits (ICs) are omnipresent, yet their globalized manufacturing process remains vulnerable to supply chain threats. Hardware Reverse Engineering (HRE) is essential for detecting such threats and re-establishing trust; however domain experts remain scarce due to a lack of educational programs. To contribute educational insights in this critical and rapidly evolving technology domain, we present our HRE course focusing on digital circuit analysis and digital circuit extraction from ICs. The course targets junior-level undergraduates at a major European research university. The curriculum has been refined over nine iterations (2017-2025), with several alumni subsequently pursuing careers in the HRE field. By reflecting on the evolution of the course organization, content, and assignments, we derive key lessons learned. We further distill these insights into actionable design priorities for educators developing courses in rapidly evolving technological domains, emphasizing iterative growth and sustainable workload management for both students and instructors.

Accurately evaluating adversarial robustness is a longstanding challenge. A flawed attack design can inflate robustness estimates, making deployment risk assessment and defense comparison unreliable. Historically, standardized attacks such as AutoAttack have largely resolved this for image classifiers, providing a reliable evaluation baseline for systematic comparison across defenses. However, no equivalent exists for LLM jailbreak evaluation yet, where designing such an attack is considerably more difficult. A reliable attack must, among other things, be black-box compatible, applicable to arbitrary defense pipelines, and efficient, which no existing method jointly satisfies. We introduce Indirect Harmfulness Optimization (IHO), a masked diffusion language model attacker trained via iterative preference optimization against a harmfulness judge, requiring only black-box access to the target. The same method can be used without modification as a strong adapting attack on individual behaviors, or as an efficient amortized policy that transfers to held-out behaviors and unseen target models without fine-tuning. Even against layered defenses, such as GLM 5.3-flash served through a black-box API deploying an auxiliary detector, IHO improves attack success considerably over state-of-the-art approaches, without any defense-specific adaptation. Our results position IHO as a practical step toward the kind of standardized jailbreak evaluation that has improved reliability in the past. Code and models are available on GitHub and Hugging Face.

Sixth-generation (6G) wireless networks will underpin ultra-dense Industrial IoT (IIoT) ecosystems in which resource-constrained Far-Edge devices – autonomous mobile robots, industrial actuators, connected vehicles – must simultaneously satisfy sub-millisecond latency, $10^{-7}$-class reliability, and decades-long cryptographic security. Current architectures delegate Digital Twin (DT) computation to centralised cloud or Mobile Edge Computing (MEC) servers, incurring prohibitive round-trip latency, and rely on classical public-key cryptography vulnerable to quantum attacks under the harvest-now, decrypt-later (HNDL) threat model. We propose Q-FE, a Quantum-Native 6G Far-Edge architecture integrating three co-designed components: (i) Micro-Digital Twins ($\mu$DTs) co-located with 6G base stations and high-capability endpoints; (ii) a Cross-Layer Post-Quantum Key Exchange module embedding CSIDH-512 isogeny key material directly within MAC-layer control frames, exploiting the scheme's uniquely compact keys ($\le 64$ bytes) to avoid packet fragmentation; and (iii) an Asynchronous Federated Learning (AFL) protocol governed by lightweight DAG smart contracts at MEC nodes, eliminating straggler bottlenecks and preventing model-poisoning and Sybil attacks without exposing raw data. End-to-end simulations (NS-3 + PySyft) demonstrate that Q-FE reduces MAC-layer overhead by 62% versus ML-KEM/Kyber-1024, maintains P99.9 URLLC latency at 0.78 ms, and accelerates global-model convergence by 31% over synchronous Federated Learning. Protocol complexity analysis confirms $O(N \log R)$ per aggregation round, and $\mu$DT handover migration completes in $1.9 \pm 0.3$ ms across $10^4$ simulated events. A formal threat model confirms resilience against quantum eavesdropping, model-poisoning, and Sybil attacks.

Large language models achieve strong performance on arithmetic reasoning benchmarks, and one common response to arithmetic brittleness is to delegate computation to code. Yet models are still often used in settings where they must reason directly from natural language, and trustworthy models should solve small-number arithmetic word problems without external tools. Prior work shows that LLMs are sensitive to numerical variation: a model may solve an original problem but fail on structurally similar variants requiring the same reasoning procedure with different numbers. We ask whether this fragility persists under a stricter setting involving small, schema-preserving numeric changes that retain the original reasoning program and avoid large-number stress tests. We introduce an automatic algorithm for generating numeric-remapping attacks on arithmetic word problems. Unlike template-based perturbation methods requiring manual schemas or constraints, our approach derives problem-specific symbolic representations, generates constrained numeric remappings, recomputes gold answers, and realizes transformed questions through deterministic edits guided by LLM-generated edit plans. Stage-wise validation and a high-confidence audit retain reliable attacks, making the pipeline scalable with limited human intervention. We evaluate DeepSeek-R1 (70B), Gemma4 (31B), and GPT-OSS (120B) on GSM8K, MAWPS, and MultiArith. On GSM8K, completed runs show conditional accuracy drops of 12.16 to 25.82 percentage points. MAWPS and MultiArith are far more stable, with most attacked accuracies near or above 98%. These results show that numeric-remapping robustness depends strongly on dataset structure: GSM8K remains sensitive even when reasoning programs are preserved and answers are recomputed, while shorter, more regular datasets are more robust.

Channel charting enables location-based services (LBSs) without requiring explicit position information by using pseudo-locations from the channel chart. While this property implies inherent privacy advantages, it does not provide formal privacy guarantees. In this work, we address location privacy in channel charting referred to as chart location indistinguishability (CLI), which extends geo-indistinguishability (GI) to channel charting representations. In order to achieve CLI, a standard planar Laplace mechanism is investigated and a geometry-aware Mahalanobis norm planar Laplace (MNPL) mechanism is devised. The proposed MNPL mechanism perturbs the channel chart by injecting noise aligned with the local structure of the chart. In the CLI framework with MNPL, privacy is defined in latent channel chart manifolds using locally adaptive covariance derived from chart neighborhoods, while preserving manifold topology under privacy constraints. In addition, differential privacy is considered as a privacy baseline. The proposed approach is evaluated across multiple channel charting schemes. The performance is assessed using utility metrics such as quality loss (QL) and range query error (RQE), as well as geometry-aware metrics including trustworthiness (TW) and continuity (CT). Numerical results demonstrate that the proposed privacy mechanism provides strong privacy guarantees while preserving the channel chart for LBSs tasks.

AI-based intrusion detection systems (IDS) have shown promise in detecting attacks on IoT systems. In this work, we explore the use of foundation models to detect and identify attacks, with a specific focus on RPL-based IoT networks. We study multiple attack types, attack variations, and network configurations, and provide insights into the performance of foundation models for attack identification. Specifically, we fine-tune the MOMENT foundation model for multi-class attack identification. Our evaluation is based on a dataset containing RPL-related statistics collected under normal operation and under Blackhole, DIS flooding, Worst Parent, and Local Repair attacks, generated in a Cooja simulation environment. The initial results are promising. The approach achieves attack-detection performance comparable to state-of-the-art methods, while also demonstrating strong performance in distinguishing between different attack types.

Early attribution of Advanced Persistent Threat (APT) activity can help defenders prioritise investigation, select countermeasures, and reduce the impact of an intrusion. Malware provides useful attribution evidence, but automated APT malware attribution remains difficult in practice. Existing approaches are typically trained and evaluated as closed-set classifiers over a limited number of known APT groups. In operational environments, however, classifiers are likely to encounter samples from groups not represented during training. Closed-set classifiers are then forced to assign such samples to known groups, producing unsupported and potentially misleading attributions. We present a high-precision APT malware attribution method based on ranked binary classifiers with explicit abstention. Rather than training a single multi-class classifier, our approach trains and tunes two binary classifiers per APT group, ranks the classifiers by validation performance, and applies them sequentially. A sample is attributed only when a classifier provides sufficient evidence; otherwise, it abstains. We evaluate the method on the APT Malware dataset and on a larger combined dataset designed to stress-test out-of-scope behaviour. On the APT Malware dataset, the method achieves higher precision than previously published results on the same dataset. In the most challenging setting, where 87% of test samples came from 60 APT groups excluded from training, the method abstained on 94% of out-of-scope samples while maintaining 92% precision and 95% selective accuracy on the samples it classified.

As AI systems evolve from passive models into autonomous active agents capable of initiating actions, collaborating, and delegating tasks, the traditional boundaries of software systems blur. Traditional authorization and delegation frameworks, built around fixed principals, explicit requests, and static scopes, are insufficient to govern agentic systems. Agentic AI demands richer authorization semantics: agents must inherit and delegate permissions, act under time-limited authority, and coordinate through shared protocols. Existing Identity and Access Management (IAM) systems fail to fully capture this notion of agency, lacking mechanisms for recursive delegation, contextual boundaries, and dynamic scoping as executable governance primitives. Unlike access delegation standards such as OAuth 2.0, we treat delegation as a contractual term rather than merely a static token-based consent credential. This paper proposes a compositional governance framework that introduces primitives indispensable for agentic AI. We define types of delegation and their permissions and accountability implications, and we introduce a notion of resource scope attenuation to bound agentic access envelopes. These concepts are expressed as general relational definitions that can be composed into existing authorization domains (e.g., financial systems). To operationalize this composition, we define a compositional operator that overlays new agentic semantics, such as recursive delegation chains, onto existing relational policies without rewriting them. We substantiate this framework through formal proofs and empirical evaluation, showing that it provides a formal yet practical foundation for accountable authorization in agentic AI systems.

With growing emphasis on privacy protection, homomorphic encryption (HE) has emerged as a core method for privacy-preserving image processing, as it enables operations directly on encrypted data. However, existing research predominantly focuses on low-resolution image processing, and techniques for privacy-preserving high-resolution image processing remain underexplored. As the image size increases, the HE parameters must be adjusted accordingly, and directly applying existing methods can lead to significant computational overhead. In this work, we propose a multi-ciphertext privacy-preserving framework for large images, enabling efficient image encryption and computation under the semi-honest model. Specifically, we divide the large image into multiple sub-images, which allows us to maintain smaller HE parameters and reduce key size. By parallel processing the sub-image ciphertexts and introducing a new bootstrapping placement strategy, we significantly reduce encryption overhead and enhance user experience. On the server side, we optimize the large image convolution operation through a repeated packing technique and implement the Sobel operator computation based on HE. To improve gradient direction calculation for the Sobel operator, we introduce a new polynomial approximation method for the reciprocal function based on the sign function, which can be applied to other HE-based protocols.

While Large Language Models (LLMs) excel in code generation, they remain prone to replicating subtle yet critical vulnerabilities endemic to their training data. Current alignment techniques, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), typically apply coarse-grained optimization at the sequence level. This approach often fails to address the localized nature of security flaws, where a single incorrect token choice can compromise an entire program. To bridge this gap, we introduce Tree-like Self-Play (TSP), a framework that reframes secure code generation as a fine-grained sequential decision process. Unlike standard methods that blindly maximize likelihood, TSP constructs a decision tree where the model explores branching trajectories–generating both secure "golden paths" and vulnerable variants. By treating code generation as a self-play game, the model learns to strictly discriminate against its own localized errors. This provides a dense, on-policy learning signal that forces self-correction precisely at the critical decision nodes where vulnerabilities typically emerge. Our experiments demonstrate that TSP fundamentally enhances model reliability. In Python security benchmarks, TSP boosts CodeLlama-7B's pass rate (SPR@1) to 75.8%, significantly outperforming SFT (57.0%) and unstructured self-play baselines. Crucially, TSP induces robust out-of-distribution generalization: the model not only reduces vulnerabilities in unseen categories (CWEs) by 24.5% but also successfully transfers security principles learned from C/C++ to diverse languages, including Python, Go, and JavaScript. This suggests that TSP does not merely memorize patches, but internalizes abstract, language-agnostic security logic.