Papers

In decentralized web applications, users face an inherent conflict between public verifiability and personal privacy. To participate in regulated on-chain services, users must currently disclose sensitive identity documents to centralized intermediaries, permanently linking real-world identities to public transaction histories. This binary choice between total privacy loss or total exclusion strips users of agency and exposes them to persistent surveillance. In this work, we introduce a Selective Disclosure Framework designed to restore user sovereignty by decoupling eligibility verification from identity revelation. We present ZK-Compliance, a prototype that leverages browser-based zero-knowledge proofs to shift the interaction model, enabling users to prove specific attributes (e.g., "I am over 18") locally without revealing the underlying data. We implement a user-governed Grant, Verify, Revoke lifecycle that transforms the user's mental model of compliance from a permanent data handover into a dynamic, revocable authorization session. Our evaluation shows that client-side proof generation takes under 200ms, enabling a seamless interactive experience on commodity hardware. This work provides early evidence that regulatory compliance need not come at the cost of user privacy or autonomy.

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.

Over the past few years an increasing number of states in the US have adopted new privacy laws. The majority of these laws require compliance with universal opt-out mechanisms (UOOMs), which allow consumers to send legally binding opt-out signals. However, a number of laws generally do not allow UOOMs to be enabled by default. While some laws exempt privacy-protective software from this prohibition, the exemption does not apply to pre-installed software, e.g., a privacy-protective web browser bundled with an operating system. The reason for not allowing default opt-out settings for pre-installed software is to ensure that settings reflect consumers' "affirmative, freely given, and unambiguous choice," as, for example, the Colorado Privacy Act (CPA) is putting it. However, prohibiting vendors of privacy-protective software from turning on UOOMs by default can force them into committing unfair or deceptive acts or practices under the FTC Act and equivalent state laws. Thus, whether UOOMs can be turned on by default on pre-installed software should depend on consumers' privacy expectations. For pre-installed software that is creating a reasonable expectation for consumers that their privacy will be protected, the simple use of such software should be considered a valid choice for enabling UOOMs. In such software a turned-on UOOM is not a "default setting" but rather the software's inherent behavior that a consumer expects and chooses through its use. This interpretation of consumer choice is preferable under the CPA and similar laws as it grounds the notice and choice principle in the privacy expectations of consumers and enables companies to compete on better privacy for consumers.

Backdoor attacks compromise model reliability by using triggers to manipulate outputs. Trigger inversion can accurately locate these triggers via a generator and is therefore critical for backdoor defense. However, the discrete nature of text prevents existing noise-based trigger generator from being applied to nature language processing (NLP). To overcome the limitations, we employ the rich knowledge embedded in large language models (LLMs) and propose a Backdoor defender powered by LLM Trigger Generator, termed BadLLM-TG. It is optimized through prompt-driven reinforcement learning, using the victim model's feedback loss as the reward signal. The generated triggers are then employed to mitigate the backdoor via adversarial training. Experiments show that our method reduces the attack success rate by 76.2% on average, outperforming the second-best defender by 13.7.

Researchers increasingly use data on social and economic networks to study a range of social science questions, but releasing statistics derived from networks can raise significant privacy concerns. We show how to release network connectedness indices that quantify assortative mixing across node attributes under edge-adjacent differential privacy. Standard privacy techniques perform poorly in this setting both because connectedness indices have high global sensitivity and because a single node's attribute can potentially be an input to connectedness in thousands of cells, leading to poor composition. Our method, which is straightforward to apply, first adds noise to node attributes, then analytically debiases downstream statistics, and finally applies a second layer of noise to protect the presence or absence of individual edges. We prove consistency and asymptotic normality of our estimators for both discrete and continuous labels and show our method works well in simulations and on real networks with as few as 200 nodes collected by social scientists.

Artificial intelligence (AI) systems are increasingly adopted as tool-using agents that can plan, observe their environment, and take actions over extended time periods. This evolution challenges current evaluation practices where the AI models are tested in restricted, fully observable settings. In this article, we argue that evaluations of AI agents are vulnerable to a well-known failure mode in computer security: malicious software that exhibits benign behavior when it detects that it is being analyzed. We point out how AI agents can infer the properties of their evaluation environment and adapt their behavior accordingly. This can lead to overly optimistic safety and robustness assessments. Drawing parallels with decades of research on malware sandbox evasion, we demonstrate that this is not a speculative concern, but rather a structural risk inherent to the evaluation of adaptive systems. Finally, we outline concrete principles for evaluating AI agents, which treat the system under test as potentially adversarial. These principles emphasize realism, variability of test conditions, and post-deployment reassessment.

Test-time training (TTT) has recently emerged as a promising method to improve the reasoning abilities of large language models (LLMs), in which the model directly learns from test data without access to labels. However, this reliance on test data also makes TTT methods vulnerable to harmful prompt injections. In this paper, we investigate safety vulnerabilities of TTT methods, where we study a representative self-consistency-based test-time learning method: test-time reinforcement learning (TTRL), a recent TTT method that improves LLM reasoning by rewarding self-consistency using majority vote as a reward signal. We show that harmful prompt injection during TTRL amplifies the model's existing behaviors, i.e., safety amplification when the base model is relatively safe, and harmfulness amplification when it is vulnerable to the injected data. In both cases, there is a decline in reasoning ability, which we refer to as the reasoning tax. We also show that TTT methods such as TTRL can be exploited adversarially using specially designed "HarmInject" prompts to force the model to answer jailbreak and reasoning queries together, resulting in stronger harmfulness amplification. Overall, our results highlight that TTT methods that enhance LLM reasoning by promoting self-consistency can lead to amplification behaviors and reasoning degradation, highlighting the need for safer TTT methods.

With the rapid development of LLM-based multi-agent systems (MAS), their significant safety and security concerns have emerged, which introduce novel risks going beyond single agents or LLMs. Despite attempts to address these issues, the existing literature lacks a cohesive safeguarding system specialized for MAS risks. In this work, we introduce TrinityGuard, a comprehensive safety evaluation and monitoring framework for LLM-based MAS, grounded in the OWASP standards. Specifically, TrinityGuard encompasses a three-tier fine-grained risk taxonomy that identifies 20 risk types, covering single-agent vulnerabilities, inter-agent communication threats, and system-level emergent hazards. Designed for scalability across various MAS structures and platforms, TrinityGuard is organized in a trinity manner, involving an MAS abstraction layer that can be adapted to any MAS structures, an evaluation layer containing risk-specific test modules, alongside runtime monitor agents coordinated by a unified LLM Judge Factory. During Evaluation, TrinityGuard executes curated attack probes to generate detailed vulnerability reports for each risk type, where monitor agents analyze structured execution traces and issue real-time alerts, enabling both pre-development evaluation and runtime monitoring. We further formalize these safety metrics and present detailed case studies across various representative MAS examples, showcasing the versatility and reliability of TrinityGuard. Overall, TrinityGuard acts as a comprehensive framework for evaluating and monitoring various risks in MAS, paving the way for further research into their safety and security.

Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks. However, they remain highly susceptible to jailbreak attacks that undermine their safety alignment. Existing defense mechanisms typically rely on post hoc filtering applied only to the final output, leaving intermediate reasoning steps unmonitored and vulnerable to adversarial manipulation. To address this gap, this paper proposes a SaFer Chain-of-Thought (SFCoT) framework, which proactively evaluates and calibrates potentially unsafe reasoning steps in real time. SFCoT incorporates a three-tier safety scoring system alongside a multi-perspective consistency verification mechanism, designed to detect potential risks throughout the reasoning process. A dynamic intervention module subsequently performs targeted calibration to redirect reasoning trajectories toward safe outcomes. Experimental results demonstrate that SFCoT reduces the attack success rate from $58.97\%$ to $12.31\%$, demonstrating it as an effective and efficient LLM safety enhancement method without a significant decline in general performance.

As telecommunications operators accelerate adoption of AI-enabled automation, a practical question remains unresolved: can general-purpose large language model (LLM) agents reliably execute telecom operations workflows through real API interfaces, or do they require structured domain guidance? We introduce SKILLS (Structured Knowledge Injection for LLM-driven Service Lifecycle operations), a benchmark framework comprising 37 telecom operations scenarios spanning 8 TM Forum Open API domains (TMF620, TMF621, TMF622, TMF628, TMF629, TMF637, TMF639, TMF724). Each scenario is grounded in live mock API servers with seeded production-representative data, MCP tool interfaces, and deterministic evaluation rubrics combining response content checks, tool-call verification, and database state assertions. We evaluate open-weight models under two conditions: baseline (generic agent with tool access but no domain guidance) and with-skill (agent augmented with a portable SKILL.md document encoding workflow logic, API patterns, and business rules). Results across 5 open-weight model conditions and 185 scenario-runs show consistent skill lift across all models. MiniMax M2.5 leads (81.1% with-skill, +13.5pp), followed by Nemotron 120B (78.4%, +18.9pp), GLM-5 Turbo (78.4%, +5.4pp), and Seed 2.0 Lite (75.7%, +18.9pp).

Mobile core networks rely on several signalling protocols in parallel, such as SS7, Diameter, and GTP, so many security-relevant problems become visible only when their interactions are analyzed jointly. At the same time, labeled examples of real attacks and cross-protocol misconfigurations are scarce, which complicates supervised detection. We therefore study unsupervised cross-protocol anomaly analysis on fused representations that combine SS7, Diameter, and GTP signalling. For each subscriber, we aggregate messages into per-minute fused records, serialize each record as text, embed it with several models, and apply unsupervised anomaly detection. We then assign each record a consensus score equal to the number of embedding models that flag it as anomalous. For evaluation, we generate cross-protocol-plausible synthetic anomalies by swapping one field group at a time between pairs of records, preserving per-message validity while making the fused view contradictory. On 219,294 fused records, 44.15% are flagged by at least one model, but only 0.97% reach full agreement across all six. Higher consensus is strongly associated with synthetic records, where for k=1-4 the odds that a flagged record is synthetic are hundreds of times greater than for original records, and for k>=5 all flagged records are synthetic, with extremely small p-values. Cosine distances between synthetic and original records also increase with consensus, suggesting clearer separation in embedding space. These results support the use of multi-embedding consensus to prioritize a much smaller set of candidate cross-protocol inconsistencies for further inspection.

An SRAM Physical Unclonable Function (PUF) can distinguish SRAM modules by analyzing the inherent randomness of their start-up behavior. However, the effectiveness of this technique varies depending on the design and fabrication of the SRAM module. This study compares two similar microcontrollers, both equipped with on-chip SRAM, to determine which device produces a better SRAM PUF. Both microcontrollers are programmed with an identical SRAM PUF authentication routine and tested under varying ambient temperatures (ranging from 10 {\deg}C to 50 {\deg}C) to evaluate the impact of temperature on SRAM PUF performance. One embedded SRAM works significantly better than the other, even though the two models are closely related. The presented results can be used early in the design process to compare arbitrary on-chip SRAM models and see which is best suited for implementing an SRAM PUF.

The safety and reliability of vision-language models (VLMs) are a crucial part of deploying trustworthy agentic AI systems. However, VLMs remain vulnerable to jailbreaking attacks that undermine their safety alignment to yield harmful outputs. In this work, we extend the Randomized Embedding Smoothing and Token Aggregation (RESTA) defense to VLMs and evaluate its performance against the JailBreakV-28K benchmark of multi-modal jailbreaking attacks. We find that RESTA is effective in reducing attack success rate over this diverse corpus of attacks, in particular, when employing directional embedding noise, where the injected noise is aligned with the original token embedding vectors. Our results demonstrate that RESTA can contribute to securing VLMs within agentic systems, as a lightweight, inference-time defense layer of an overall security framework.

In cloud-based endpoint auditing, security administrators often rely on the cloud to perform causality analysis over log-derived versioned provenance graphs to investigate suspicious attack behaviors. However, the cloud may be distrusted or compromised by attackers, potentially manipulating the final causality analysis results. Consequently, administrators may not accurately understand attack behaviors and fail to implement effective countermeasures. This risk underscores the need for a defense scheme to ensure the integrity of causality analysis. While existing tamper-evident logging schemes and trusted execution environments show promise for this task, they are not specifically designed to support causality analysis and thus face inherent security and efficiency limitations. This paper presents vCause, an efficient and verifiable causality analysis system for cloud-based endpoint auditing. vCause integrates two authenticated data structures: a graph accumulator and a verifiable provenance graph. The data structures enable validation of two critical steps in causality analysis: (i) querying a point-of-interest node on a versioned provenance graph, and (ii) identifying its causally related components. Formal security analysis and experimental evaluation show that vCause can achieve secure and verifiable causality analysis with only <1% computational overhead on endpoints and 3.36% on the cloud.

We study trivariate permutation polynomials over $\mathbb{F}_{2^{m}}$ extending two APN permutation families of Li–Kaleyski (IEEE Trans. Inform. Theory, 2024) by allowing the scalar parameter to vary over $\mathbb{F}_{2^m}^*$. For \[ G_a(x,y,z)=(x^{q+1}+ax^qz+yz^q,\; x^qz+y^{q+1},\; xy^q+ay^qz+z^{q+1}), \] where $a\in\mathbb{F}_{2^m}^*$, $q=2^i$, $\gcd(i,m)=1$, and $m$ is odd, we prove that $G_a$ is a permutation if and only if an associated univariate polynomial has no root in $\mathbb{F}_{2^m}^*$, and that this condition is also equivalent to $G_a$ being APN. Hence, writing $d=q^2+q+1$, at least \[ \frac{2^m+1-(d-1)(d-2)2^{m/2}-d}{d} \] values of $a$ yield APN permutations $G_a$. In the binary case $q=2$, we show that $a=1$ is good whenever $7\nmid m$, recovering the Li–Kaleyski family. For the second family \[ H_a(x,y,z)=(x^{q+1}+axy^q+yz^q,\; xy^q+z^{q+1},\; x^qz+y^{q+1}+ay^qz), \] we obtain the same root criterion and prove that its defining polynomial is root-equivalent to that of $G_a$. Thus the same parameters $a$ give APN permutations in both families. We also prove strong inequivalence results. First, $G_a$ (resp.\ $H_a$) is diagonally equivalent to $G_1$ (resp.\ $H_1$) if and only if $a^{q^2+q+1}=1$; moreover, for $m>4$, $m\neq 6$, and $7\nmid m$, diagonal non-equivalence implies CCZ non-equivalence by the monomial restriction theorem of Shi et al.\ (DCC, 2025). In particular, when $q=2$ and $7\nmid m$, every good $a\neq 1$ gives APN permutations CCZ-inequivalent to Li–Kaleyski. Second, for the same range of $m$, no $G_a$ is CCZ-equivalent to any $H_b$. Hence these constructions yield two genuinely new, mutually inequivalent families of APN permutations on $\mathbb{F}_{2^{3m}}$.

Modern agentic systems allow Large Language Model (LLM) agents to tackle complex tasks through extensive tool usage, forming structured control flows of tool selection and execution. Existing security analyses often treat these control flows as ephemeral, one-off sessions, overlooking the persistent influence of memory. This paper identifies a new threat from Memory Control Flow Attacks (MCFA) that memory can dominate the control flow, forcing unintended tool usage even against explicit user instructions and inducing persistent behavioral deviations across tasks. To understand the impact of this vulnerability, we further design MEMFLOW, an automated evaluation framework that systematically identifies and quantifies MCFA across heterogeneous tasks and long interaction horizons. To evaluate MEMFLOW, we attack state-of-the-art LLMs, including GPT-5 mini, Claude Sonnet 4.5 and Gemini 2.5 Flash on real-world tools from two major LLM agent development frameworks, LangChain and LlamaIndex. The results show that in general over 90% of trials are vulnerable to MCFA even under strict safety constraints, highlighting critical security risks that demand immediate attention.

Answering Select-Join-Aggregate queries with DP is a fundamental problem with important applications in various domains. The current SOTA methods ensure user-level DP (i.e., the adversary cannot infer the presence or absence of any given individual user with high confidence) and achieve instance-optimal accuracy on the query results. However, these solutions involve solving expensive optimization programs, which may incur prohibitive computational overhead for large databases. One promising direction to achieve scalability is through sampling, which provides a tunable trade-off between result utility and computational costs. However, applying sampling to differentially private SJA processing is a challenge for two reasons. First, it is unclear what to sample, in order to achieve the best accuracy within a given computational budget. Second, prior solutions were not designed with sampling in mind, and their mathematical tool chains are not sampling-friendly. To our knowledge, the only known solution that applies sampling to private SJA processing is S&E, a recent proposal that (i) samples users and (ii) combines sampling directly with existing solutions to enforce DP. We show that both are suboptimal designs; consequently, even with a relatively high sample rate, the error incurred by S&E can be 10x higher than the underlying DP mechanism without sampling. Motivated by this, we propose Differentially Private Sampling for Scale (DP-S4S), a novel mechanism that addresses the above challenges by (i) sampling aggregation units instead of users, and (ii) laying the mathematical foundation for SJA processing under RDP, which composes more easily with sampling. Further, DP-S4S can answer both scalar and vector SJA queries. Extensive experiments on real data demonstrate that DP-S4S enables scalable SJA processing on large datasets under user-level DP, while maintaining high result utility.

While watermarking serves as a critical mechanism for LLM provenance, existing secret-key schemes tightly couple detection with injection, requiring access to keys or provider-side scheme-specific detectors for verification. This dependency creates a fundamental barrier for real-world governance, as independent auditing becomes impossible without compromising model security or relying on the opaque claims of service providers. To resolve this dilemma, we introduce TTP-Detect, a pioneering black-box framework designed for non-intrusive, third-party watermark verification. By decoupling detection from injection, TTP-Detect reframes verification as a relative hypothesis testing problem. It employs a proxy model to amplify watermark-relevant signals and a suite of complementary relative measurements to assess the alignment of the query text with watermarked distributions. Extensive experiments across representative watermarking schemes, datasets and models demonstrate that TTP-Detect achieves superior detection performance and robustness against diverse attacks.

We present a fine-tuned RoBERTa-base classifier (125M parameters) for mapping Common Vulnerabilities and Exposures (CVE) descriptions to Common Weakness Enumeration (CWE) categories. We construct a large-scale training dataset of 234,770 CVE descriptions with AI-refined CWE labels using Claude Sonnet 4.6, and agreement-filtered evaluation sets where NVD and AI labels agree. On our held-out test set (27,780 samples, 205 CWE classes), the model achieves 87.4% top-1 accuracy and 60.7% Macro F1 – a +15.5 percentage-point Macro F1 gain over a TF-IDF baseline that already reaches 84.9% top-1, demonstrating the model's advantage on rare weakness categories. On the external CTI-Bench benchmark (NeurIPS 2024), the model achieves 75.6% strict accuracy (95% CI: 72.8-78.2%) – statistically indistinguishable from Cisco Foundation-Sec-8B-Reasoning (75.3%, 8B parameters) at 64x fewer parameters. We release the dataset, model, and training code.

Generative AI deployment poses unprecedented challenges to content safety and privacy. However, existing defense mechanisms are often tailored to specific architectures (e.g., Diffusion Models or GANs), creating fragile "defense silos" that fail against heterogeneous generative threats. This paper identifies a fundamental optimization barrier in naive pixel-space ensemble strategies: due to divergent objective functions, pixel-level gradients from heterogeneous generators become statistically orthogonal, causing destructive interference. To overcome this, we observe that despite disparate low-level mechanisms, high-level feature representations of generated content exhibit alignment across architectures. Based on this, we propose the Architecture-Agnostic Targeted Feature Synergy (ATFS) framework. By introducing a target guidance image, ATFS reformulates multi-model defense as a unified feature space alignment task, enabling intrinsic gradient alignment without complex rectification. Extensive experiments show ATFS achieves SOTA protection in heterogeneous scenarios (e.g., Diffusion+GAN). It converges rapidly, reaching over 90% performance within 40 iterations, and maintains strong attack potency even under tight perturbation budgets. The framework seamlessly extends to unseen architectures (e.g., VQ-VAE) by switching the feature extractor, and demonstrates robust resistance to JPEG compression and scaling. Being computationally efficient and lightweight, ATFS offers a viable pathway to dismantle defense silos and enable universal generative security. Code and models are open-sourced for reproducibility.

Quantum computing provides a feasible multi-layered security challenge to classical blockchain networks. Quantum blockchains that rely on quantum key distribution (QKD) to establish secure channels can address this feasible threat. Whereas, there are still architecture limitations to practical security resulted in the measurement devices while implementing the QKD-secured blockchains in physical layer. This paper presents a quantum-classical hybrid architecture in a distributed blockchain to address the connectivity and distance limitations of the blockchain-embedded quantum networks. A decoupled architecture is designed felicitously so that it pairs a linearly scalable measurement-device-independent (MDI) physical layer with a decentralized consensus. It can optimize the complexity of infrastructure from quadratic to linear scaling, ascribed to leveraging the twin-field (TF) QKD protocol with the MDI-structurized star topology. Additionally, the dual-key stratification strategy transforms symmetric information-theoretic security into publicly auditable forward-secret blockchain evidence. This architecture can integrate the exact information-theoretic security (ITS) with distributed consensus mechanisms, allowing the scalable system to overcome the potential rate-loss limits inherent in classical security-weakened blockchains.

Retrieval-Augmented Generation (RAG) enables large language models to use external knowledge, but outsourcing the RAG service raises privacy concerns for both data owners and users. Privacy-preserving RAG systems address these concerns by performing secure top-$k$ retrieval, which is typically implemented using secure sorting to identify relevant documents. However, existing systems face challenges supporting arbitrary $k$ due to their inability to change $k$, new security issues, and in particular, efficiency degradation with large $k$. This is a significant limitation because applications such as finance, law, and healthcare require a $k$ that is large enough to cause huge overhead for existing systems. Also, modern long-context models generally achieve higher accuracy with larger retrieval sets. We propose P$^2$RAG, an efficient privacy-preserving RAG service that supports arbitrary top-$k$ retrieval. Unlike existing systems, P$^2$RAG avoids sorting candidate documents. Instead, it uses an interactive bisection method to determine the set of top-$k$ documents. For security, P$^2$RAG uses secret sharing on two semi-honest non-colluding servers to protect the data owner's database and the user's prompt. It enforces restrictions and verification to defend against malicious users and tightly bounds the information leakage of the database. The experiments show that P$^2$RAG is 3–300$\times$ faster than the state-of-the-art PRAG for $k = 16$–$1024$.

Canonical LST (sTEZ) is an enshrined, protocol-native mechanism designed to mitigate the centralization risks associated with liquid staking intermediaries. Intended to complement direct staking rather than replace it, Canonical LST provides a neutral, public alternative managed directly by the Tezos protocol. It allows any tez holder to participate in aggregated staking without reliance on third-party operators. sTEZ follows an accrual-based design: all slashing events and rewards are reflected in the token's exchange rate to tez, keeping balances fungible while exposing holders to the precise economics of staking. This approach ensures that liquid staking functions as fundamental network infrastructure–with deterministic lifecycle rules, transparent on-chain data, and governance anchored in the amendment process–rather than as a discretionary commercial product. This white paper summarises the motivation for enshrining liquid staking, the core mechanics, exchange-rate model, regulatory touchpoints, risk posture, and forward-looking roadmap for Canonical LST.

Contemporary AI governance frameworks rely heavily on post hoc oversight, policy guidance, and behavioral alignment techniques, yet these mechanisms become fragile as systems gain autonomy, speed, and operational opacity. This paper presents Aegis, a runtime governance architecture for autonomous AI systems that treats policy and legal constraints as execution conditions rather than advisory principles. Aegis binds each governed agent to a cryptographically sealed Immutable Ethics Policy Layer (IEPL) at system genesis and enforces external emissions through an Ethics Verification Agent (EVA), an Enforcement Kernel Module (EKM), and an Immutable Logging Kernel (ILK). Amendments to the governing policy layer require quorum approval and redeclaration of the system trust root; verified violations trigger autonomous shutdown and generation of auditable proof artifacts. We evaluate the architecture within the Civitas runtime using three operational measures: proof verification latency under tamper conditions, publication overhead, and alignment retention performance relative to an ungoverned baseline. In controlled trials, Aegis demonstrates median proof verification latency of 238 ms, median publication overhead of approximately 9.4 ms, and higher alignment retention than the baseline condition across matched tasks. We argue that these results support a shift in AI governance from discretionary oversight toward verifiable runtime constraint. Rather than claiming to resolve machine ethics in the abstract, the proposed architecture seeks to show that policy violating behavior can be rendered operationally non executable within a controlled runtime governance framework. The paper concludes by discussing methodological limits, evidentiary implications, and the role of proof oriented governance in high assurance AI deployment.

Safety alignment in large language models is typically evaluated under isolated queries, yet real-world use is inherently multi-turn. Although multi-turn jailbreaks are empirically effective, the structure of conversational safety failure remains insufficiently understood. In this work, we study safety failures from a state-space perspective and show that many multi-turn safety failures in current safety-aligned language models arise from contextual state evolution, a regime that is not fully captured by isolated prompt-level analyses alone. We introduce STAR, a state-oriented diagnostic framework that treats dialogue history as a state transition operator and enables controlled analysis of safety behavior along interaction trajectories. Rather than optimizing attack strength, STAR provides a principled probe of how aligned models traverse the safety boundary under autoregressive conditioning. Across multiple frontier language models, we find that systems that appear robust under static evaluation can undergo rapid and reproducible safety collapse under structured multi-turn interaction. Mechanistic analysis reveals monotonic drift away from refusal-related representations and abrupt phase transitions induced by role-conditioned context. Together, these findings motivate viewing language model safety as a dynamic, state-dependent process defined over conversational trajectories.

Automated LLM vulnerability scanners are increasingly used to assess security risks by measuring different attack type success rates (ASR). Yet the validity of these measurements hinges on an often-overlooked component: the evaluator who determines whether an attack has succeeded. In this study, we demonstrate that commonly used open-source scanners exhibit measurement instability that depends on the evaluator component. Consequently, changing the evaluator while keeping the attacks and model outputs constant can significantly alter the reported ASR. To tackle this problem, we present a two-phase, reliability-aware evaluation framework. In the first phase, we quantify evaluator disagreement to identify attack categories where ASR reliability cannot be assumed. In the second phase, we propose a verification-based evaluation method where evaluators are validated by an independent verifier, enabling reliability assessment without relying on extensive human annotation. Applied to the widely used Garak scanner, we observe that 22 of 25 attack categories exhibit evaluator instability, reflected in high disagreement among evaluators. Our approach raises evaluator accuracy from 72% to 89% while enabling selective deployment to control cost and computational overhead. We further quantify evaluator uncertainty in ASR estimates, showing that reported vulnerability scores can vary by up to 33% depending on the evaluator. Our results indicate that the outputs of vulnerability scanners are highly sensitive to the choice of evaluators. Our framework offers a practical approach to quantify unreliable evaluations and enhance the reliability of measurements in automated LLM security assessments.

Neurosymbolic approaches leveraging Large Language Models (LLMs) with formal methods have recently achieved strong results on mathematics-oriented theorem-proving benchmarks. However, success on competition-style mathematics does not by itself demonstrate the ability to construct proofs about real-world implementations. We address this gap with a benchmark derived from an industrial cryptographic library whose assembly routines are already verified in HOL Light. s2n-bignum is a library used at AWS for providing fast assembly routines for cryptography, and its correctness is established by formal verification. The task of formally verifying this library has been a significant achievement for the Automated Reasoning Group. It involved two tasks: (1) precisely specifying the correct behavior of a program as a mathematical proposition, and (2) proving that the proposition is correct. In the case of s2n-bignum, both tasks were carried out by human experts. In s2n-bignum-bench, we provide the formal specification and ask the LLM to generate a proof script that is accepted by HOL Light within a fixed proof-check timeout. To our knowledge, s2n-bignum-bench is the first public benchmark focused on machine-checkable proof synthesis for industrial low-level cryptographic assembly routines in HOL Light. This benchmark provides a challenging and practically relevant testbed for evaluating LLM-based theorem proving beyond competition mathematics. The code to set up and use the benchmark is available here: \href{https://github.com/kings-crown/s2n-bignum-bench}{s2n-bignum-bench}.

Financial fraud detection in transaction networks involves modeling sparse anomalies, dynamic patterns, and severe class imbalance in the presence of temporal drift in the data. In real-world transaction systems, a suspicious transaction is rarely isolated: rather, legitimate and suspicious transactions are often connected through accounts, intermediaries or through temporal transaction sequences. Attribute-based or randomly partitioned learning pipelines are therefore insufficient to detect relationally structured fraud. STC-MixHop, a graph-based framework combining spatial multi-resolution propagation with lightweight temporal consistency modeling for anomaly and fraud detection in dynamic transaction networks. It integrates three components: a MixHop-inspired multi-scale neighborhood diffusion encoder a multi-scale neighborhood diffusion MixHop-based encoder for learning structural patterns; a spatial-temporal attention module coupling current and preceding graph snapshots to stabilize representations; and a temporally informed self-supervised pretraining strategy exploiting unlabeled transaction interactions to improve representation quality. We evaluate the framework primarily on the PaySim dataset under strict chronological splits, supplementing the analysis with Porto Seguro and FEMA data to probe cross-domain component behavior. Results show that STC-MixHop is competitive among graph methods and achieves strong screening-oriented recall under highly imbalanced conditions. The experiments also reveal an important boundary condition: when node attributes are highly informative, tabular baselines remain difficult to outperform. Graph structure contributes most clearly where hidden relational dependencies are operationally important. These findings support a stability-focused view of graph learning for financial fraud detection.

As database deployments shift toward cloud platforms and edge devices, thin clients need to securely retrieve sensitive records without leaking their query intent or metadata to the proxies that mediate access. Oblivious Transfer (OT) is a core tool for private retrieval, yet existing OTs assume direct client-database interaction and lack support for delegated querying or lightweight clients. We present Oblivis, a modular framework of new OT protocols that enable delegated, privacy-preserving query execution. Oblivis allows clients to retrieve database records without direct access, protects against leakage to both databases and proxies, and is designed with practical efficiency in mind. Its components include: (1) Delegated-Query OT, which permits secure outsourcing of query generation; (2) Multi-Receiver OT for merged, cloud-hosted databases; (3) a compiler producing constant-size responses suitable for thin clients; and (4) Supersonic OT, a proxy-based, informationtheoretic, and highly efficient 1-out-of-2 OT. The protocols are formally defined and proven secure in the simulation-based paradigm, under non-colluding assumption. We implement and empirically evaluate Supersonic OT. It achieves at least a 92x speedup over a highly efficient 1-out-of-2 OT, and a 2.6x-106x speedup over a standard OT extension across 200-100,000 invocations. Our implementation further shows that Supersonic OT remains efficient even on constrained hardware, e.g., it completes an end-to-end transfer in 1.36 ms on a Raspberry Pi 4.

Electronic banking portals often sit in front of enterprise resource planning (ERP) systems such as SAP, mediating payment requests between users and back end financial infrastructure. When these integrations place excessive trust in client supplied HTTP metadata, subtle design flaws can arise that undermine payment integrity. This article presents a retrospective, anonymized case study of an SAP based payment flow in which weaknesses in HTTP level validation allowed the front end application to incorrectly treat unpaid transactions as completed. Rather than provide a reproducible exploit, we abstract the scenario into a general vulnerability pattern, analyze contributing architectural decisions, and propose concrete design and verification practices for secure web to ERP payment processing. The discussion emphasizes formalizing payment state machines, strengthening trust boundaries, and incorporating regular security review into integration projects.