Watermarking has recently emerged as a crucial tool for protecting the intellectual property of generative models and for distinguishing AI-generated content from human-generated data. Despite its practical success, most existing watermarking schemes are empirically driven and lack a theoretical understanding of the fundamental trade-off between detection power and generation fidelity. To address this gap, we formulate watermarking as a statistical hypothesis testing problem between a null distribution and its watermarked counterpart. Under explicit constraints on false-positive and false-negative rates, we derive a tight lower bound on the achievable fidelity loss, measured by a general f-divergence, and characterize the optimal watermarked distribution that attains this bound. We further develop a corresponding sampling rule that provides an optimal mechanism for inserting watermarks with minimal fidelity distortion. Our result establishes a simple yet broadly applicable principle linking hypothesis testing, information divergence, and watermark generation.

Generative AI makes social-engineering attacks more fluent, adaptive, and scalable, increasing the need for LLM-based de- fenders that can protect users during ongoing interactions. We ask whether such defenders identify the structural source of risk or merely react to surface cues. We formalize trust-chain localization: identifying whether an interaction fails at actor authority, asset control, verification sufficiency, or transaction path. We construct a controlled 300-case online-housing corpus spanning 20 scenario families, legitimate cases, four structural failure modes, and three surface conditions. Five defender models are evaluated on the same corpus in state- ful turn-by-turn and one-shot static settings, yielding 1,500 model-case evaluations per protocol and 3,000 in total. No model produced explicit unsafe compliance, yet defensive effectiveness varied sharply: intervention rates ranged from 0

On August 2, 2026, the obligations of Article 50 of the EU AI Act took effect, requiring generative AI providers to mark the content their systems produce and ensure it can be detected as AI-generated. Days later, Anthropic disclosed that every Claude model released after that date embeds a watermark based on SynthID-Text in all generated text, enabled by default with no user opt-out; Google has deployed SynthID-Text in Gemini since 2024. Users objected that the watermark degrades quality, particularly for code, that it secretly encodes identifying information, and, in mutual contradiction, that it is easily removable and inescapable; the vendor answered with assurances of unchanged quality, no identifying information, and robustness to light editing. In this work, we argue that neither the objections nor the assurances can currently be verified and that this unverifiability, rather than watermarking itself, is the substantive governance failure. We sort the contested assertions by what it would take to settle each and evaluate the open-source SynthID-Text implementation on two open-weight models, because no public tool can test the deployed systems. On prose, the measured effect of the watermark does not exceed that of changing the sampling seed. On code, the cost is three points of correctness on one model and below measurement on the other, while detection remains near chance, a limitation of detectability rather than quality. The remaining gaps trace to withheld access or missing institutions and we map each to a requirement: release of matched outputs, configuration disclosure, accredited audits, a shared evaluation protocol, and interoperable detection.

Standard PII filters often miss contextual data leakage in RAG systems, such as non-regulated attribute clusters that collectively identify individuals. We introduce a Privacy Policy Enforcement (PPE) framework using dual one-class density estimators with fused text embeddings and a calibrated abstain region for out-of-distribution inputs. Using an axis-stratified, multi-LLM synthetic data pipeline across medicine, finance, and law, we found that traditional Gaussian Mixture baselines fail on borderline-safe stress tests by focusing on linguistic register rather than content. Our proposed T3+OCSVM detector, trained on safe and borderline-safe data, achieves a borderline AUROC of 0.93+ while reducing false positives by 44-55 percentage points and maintaining millisecond latency. Compared to supervised MLP classifiers or 14B-parameter LLM judges, our framework offers superior operational suitability, as the former suffers from high abstention rates and the latter from latency and calibration issues. This methodology provides a robust stress-testing standard for any synthetic-data-trained classifier.

In recent years, the widespread adoption of Machine Learning as a Service (MLaaS), particularly in sensitive environments, has raised considerable privacy concerns. Of particular importance are membership inference attacks (MIAs), which exploit behavioral discrepancies between training and non-training data to determine whether a specific record was included in the model's training set, thereby presenting significant privacy risks. Although existing defenses, such as adversarial regularization, DP-SGD, and MemGuard, assist in mitigating these threats, they often entail trade-offs such as compromising utility, increased computational requirements, or inconsistent protection against diverse attack vectors. In this paper, we introduce a novel inference-time defense mechanism called Neighborhood Blending, which mitigates MIAs without retraining the model or incurring significant computational overhead. Our approach operates post-training by smoothing the model's confidence outputs based on the neighborhood of a queried sample. By averaging predictions from similar training samples selected using differentially private sampling, our method establishes a consistent confidence pattern, rendering members and non-members indistinguishable to an adversary while maintaining high utility. Significantly, Neighborhood Blending maintains label integrity (zero label loss) and ensures high utility through an adaptive, "pay-as-you-go" distortion strategy. It is a model-agnostic approach that offers a practical, lightweight solution that enhances privacy without sacrificing model utility. Through extensive experiments across diverse datasets and models, we demonstrate that our defense significantly reduces MIA success rates while preserving model performance, outperforming existing post-hoc defenses like MemGuard and training-time techniques like DP-SGD in terms of utility retention.

Estimating treatment effects from observational data requires choosing an adjustment set, but valid adjustment depends on an unknown causal graph. Graph misspecification can cause under-coverage, while graph-agnostic conformal wrappers may regain nominal coverage only through large padding. We introduce CausalGuard, a structure-weighted conformal framework that calibrates after aggregating graph-conditional doubly robust pseudo-outcomes. Candidate DAGs are proposed from an LLM-derived edge prior, pruned by conditional-independence tests, and reweighted by Bayesian Information Criterion. A composite nonconformity score then calibrates the posterior-weighted pseudo-outcome. CausalGuard provides distribution-free finite-sample marginal coverage for this aggregated pseudo-outcome; under causal identification, overlap, conditional-mean nuisance stability, and concentration on target-aligned valid adjustment strategies, its conditional mean converges to the true Conditional Average Treatment Effect. Across five benchmarks, CausalGuard attains mean coverage above the nominal 90

Genomic language models (GLMs) have emerged as powerful tools for learning representations of DNA sequences, enabling advances in variant prediction, regulatory element identification, and cross-task transfer learning. However, as these models are increasingly trained or fine-tuned on sensitive genomic cohorts, they risk memorizing specific sequences from their training data, raising serious concerns around privacy, data leakage, and regulatory compliance. Despite growing awareness of memorization risks in general-purpose language models, little systematic evaluation exists for these risks in the genomic domain, where data exhibit unique properties such as a fixed nucleotide alphabet, strong biological structure, and individual identifiability. We present a comprehensive, multi-vector privacy evaluation framework designed to quantify memorization risks in GLMs. Our approach integrates three complementary risk assessment methodologies: perplexity-based detection, canary sequence extraction, and membership inference. These are combined into a unified evaluation pipeline that produces a worst-case memorization risk score. To enable controlled evaluation, we plant canary sequences at varying repetition rates into both synthetic and real genomic datasets, allowing precise quantification of how repetition and training dynamics influence memorization. We evaluate our framework across multiple GLM architectures, examining the relationship between sequence repetition, model capacity, and memorization risk. Our results establish that GLMs exhibit measurable memorization and that the degree of memorization varies across architectures and training regimes. These findings reveal that no single attack vector captures the full scope of memorization risk, underscoring the need for multi-vector privacy auditing as a standard practice for genomic AI systems.

The data revolution presents unprecedented potential for advancements in the healthcare sector, particularly within genomic research. On one hand, the continuous accumulation of large-scale individual genomic data serves as the foundation for the development of artificial intelligence-based innovations and digital health technologies. On the other hand, genomic data raises unique security and privacy risks requiring structured threat modeling and privacy-preserving measures. This work presents an implementation of a privacy-preserving framework specifically designed for genomic research, integrated into a real-world secure platform for medical data collaboration. The proposed framework addresses critical security and privacy vulnerabilities, enabling the controlled sharing and analysis of genomic datasets while mitigating risks associated with data breaches. By leveraging advanced privacy-preserving algorithms, the framework protects privacy while maintaining data utility. A key aspect of this approach is the strategic trade-offs between data sharing and privacy, providing stakeholders with quantifiable metrics to assess privacy risks and inform data-sharing decisions. The implementation within a real-world platform encompasses encoding genomic data into binary formats and introducing controlled noise to preserve key statistical attributes. This ensures the integrity of research outcomes while enabling privacy-aware computational analyses. Additionally, the envisioned framework integrates real-time data monitoring and advanced visualization tools, optimizing user experience and decision-making processes. Given the unique characteristics of genomic data, our work underscores the necessity for tailored privacy attacks and corresponding defenses to safeguard sensitive information effectively. By addressing these challenges, the proposed solution aspires to foster a global research ecosystem in genomics, ultimately accelerating breakthroughs in personalized medicine and public health.

As genomic research has grown increasingly popular in recent years, dataset sharing has remained limited due to privacy concerns. This limitation hinders the reproducibility and validation of research outcomes, both of which are essential for identifying computational errors during the research process. In this paper, we introduce PROVGEN, a privacy-preserving method for sharing genomic datasets that facilitates reproducibility and outcome validation in genome-wide association studies (GWAS). Our approach encodes genomic data into binary space and applies a two-stage process. First, we generate a differentially private version of the dataset using an XOR-based mechanism tailored to biological characteristics. Second, we restore data utility by adjusting the Minor Allele Frequency (MAF) values in the noisy dataset to align with public MAFs using optimal transport. Finally, we convert the processed binary data back into its genomic representation and publish the resulting dataset. We evaluate PROVGEN on three real-world genomic datasets and compare it with local differential privacy and three synthesis-based methods. Our results show that PROVGEN overall outperforms existing approaches in detecting GWAS outcome errors, preserving data fidelity, and resisting membership inference attacks (MIAs). By adopting our method, genomic researchers will be inclined to share differentially private datasets while maintaining high data quality for reproducibility of their findings.

Modern healthcare analytics increasingly relies on complex AI-driven workflows operating over sensitive datasets across institutional boundaries. Ensuring secure, auditable, and policy-compliant execution of these workflows remains a significant challenge. Existing systems typically treat workflow generation, policy enforcement, and execution auditing as separate concerns, resulting in fragmented governance mechanisms. This article proposes a governance-aware workflow lifecycle architecture for AI-driven analytics workflows that integrates natural language workflow synthesis, blockchain-based policy enforcement, and auditable execution, embedding governance directly within the workflow lifecycle. In our framework, large language models translate user requests into executable analytics workflows represented using Business Process Model and Notation (BPMN). Access control policies defined in XACML are compiled into smart contracts that enforce policy compliance and record immutable audit logs on a blockchain. Sensitive datasets remain off-chain while metadata and compliance events are securely recorded. We implement a prototype platform for healthcare analytics demonstrating the generation and execution of survival analysis workflows. We further evaluate the robustness of LLM-generated workflows through an ablation study examining how prompt structure affects BPMN diagram correctness and completeness. Results highlight both the promise and limitations of automated workflow generation in regulated analytics environments. Our work demonstrates the feasibility of integrating governance mechanisms directly into AI workflow infrastructures, providing a foundation for secure and auditable analytics in sensitive domains.

For decades, the security of digital interaction has rested on an unacknowledged economic constraint. Attackers faced a tradeoff between the fidelity of a deception and the scale at which it could be deployed. Convincing impersonation required sustained human effort and was confined to a narrow set of high-value targets, while mass-market attacks sacrificed plausibility for reach. Detection systems, verification mechanisms, and user awareness training have all been implicitly calibrated to the artifacts of cheap deception that this tradeoff produced. Agentic AI collapses the tradeoff, allowing high-fidelity, individually tailored deception to be produced at mass-market scale. We argue that this shift exhausts a security paradigm rather than merely intensifying the threat landscape. We introduce the Infinite Impostor, an attack model in which an autonomous agent interposes itself between two parties who already trust each other, hijacking an existing relationship rather than building a new one from scratch. Detection-oriented defenses share an assumption that generative progress is eliminating, that synthetic outputs are distinguishable from authentic ones. We propose a suspect-by-default paradigm that shifts security from authenticating actors to evaluating actions, and examine the governance tensions that arise when platforms become the regulatory substrate of digital interaction.

Cryptographic provenance standards such as C2PA and invisible watermarking are positioned as complementary defenses for content authentication, yet the two verification layers are technically independent: neither conditions on the output of the other. This work formalizes and empirically demonstrates the Integrity Clash, a condition in which a digital asset carries a cryptographically valid C2PA manifest asserting human authorship while its pixels simultaneously carry a watermark identifying it as AI-generated, with both signals passing their respective verification checks in isolation. We construct metadata washing workflows that produce these authenticated fakes through standard editing pipelines, requiring no cryptographic compromise, only the semantic omission of a single assertion field permitted by the current C2PA specification. To close this gap, we propose a cross-layer audit protocol that jointly evaluates provenance metadata and watermark detection status, achieving 100

Motivation Genomic data is among the most sensitive categories of personal information, and the growing adoption of language models for sequence analysis raises significant privacy concerns. Prior work demonstrated that embeddings from general-purpose language models adapted for genomic sequences leak substantial single-nucleotide information under reconstruction attacks, and that fine-tuning embeddings can reduce this vulnerability at certain positions. However, three critical questions remain unaddressed: (i) whether privacy-utility tradeoffs are inherent constraints or configuration-dependent phenomena; (ii) whether genomic-specialized models such as DNABERT-base and Nucleotide Transformer exhibit different vulnerabilities than adapted general-purpose models; and (iii) how to statistically validate whether observed privacy improvements represent meaningful gains. Addressing these gaps is essential for guiding model selection in privacy-sensitive genomic applications.Results We systematically evaluated 13 transformer architectures, 9 general-purpose and 4 genomic-specialized, under position-specific embedding reconstruction attacks. We assessed the vulnerabilities of both pre-trained and fine-tuned models to the single-nucleotide inference-reconstruction attack using our new metrics, including error-based privacy gain and Pareto dominance scores, and statistically validated the results via paired t-tests. XLNet-Large achieved the best observed privacy protection among all evaluated models (+19.5% mean privacy gain) while maintaining competitive prediction performance. General-purpose models outperformed genomic-specialized models in 56% of pairwise comparisons. Tokenization strategy, rather than domain specialization, emerged as the primary determinant of the privacy-utility balance. These findings provide evidence-based guidance for selecting models in privacy-sensitive short-window genomic applications. All privacy claims in this work are specific to position-wise embedding reconstruction attacks and do not extend to other privacy risks, such as membership inference or training data extraction, which may respond differently to fine-tuning.Availability and implementation The code is publicly available at https://github.com/AnonymousISCBConf/Win-Win-Privacy-Utility-Analysis.

Objectives:To verify that federated genomic study sites applied identical preprocessing pipelines without disclosing raw genotypes. Materials and Methods:Each institution perturbs a 100-SNP slice using local differential privacy (LDP), trains a RandomForest classifier, and transmits one LIME explanation vector to a coordinating server. The server simulates 15 preprocessing combinations and trains a RandomForest classifier to predict each site's configuration. Results:In centralized simulation, the verifier achieved 80% accuracy across 15 preprocessing configurations on the GMMAT (n = 400) and synthetic genome (n = 2504) datasets while maintaining membership-inference attack power below 0.05 at ε = 3. In distributed Flower FL experiments with data partitioned across three sites, binary compatibility detection reached 70% accuracy at 500 SNPs. Discussion:A single differentially private explanation vector provides an auditable preprocessing fingerprint. The gap between centralized and distributed accuracy reflects expected FL data partitioning effects. Conclusion:This framework demonstrates the feasibility of automated preprocessing verification in federated genomic consortia without compromising participant privacy.

The indistinguishability of large language model (LLM) output from human-authored content poses significant challenges, raising concerns about potential misuse of AI-generated text and its influence on future model training. Watermarking algorithms offer a viable solution by embedding detectable signatures into generated text. However, existing watermarking methods often involve trade-offs among attack robustness, generation quality, and additional overhead such as specialized frameworks or complex integrations. We propose a lightweight, topic-guided watermarking scheme for LLMs that partitions the vocabulary into topic-aligned token subsets. Given an input prompt, the scheme selects a relevant topic-specific token list, effectively "green-listing" semantically aligned tokens to embed robust marks while preserving fluency and coherence. Experimental results across multiple LLMs and state-of-the-art benchmarks demonstrate that our method achieves text quality comparable to industry-leading systems and simultaneously improves watermark robustness against paraphrasing and lexical perturbation attacks, with minimal performance overhead. Our approach avoids reliance on additional mechanisms beyond standard text generation pipelines, enabling straightforward adoption and suggesting a practical path toward globally consistent watermarking of AI-generated content.

Watermarking is becoming the default mechanism for AI content authentication, with governance policies and frameworks referencing it as infrastructure for content provenance. Yet across text, image, and audio modalities, watermark signal strength, detectability, and robustness depend on statistical properties of the content itself, properties that vary systematically across languages, cultural visual traditions, and demographic groups. We examine how this content dependence creates modality-specific pathways to bias. Reviewing the major watermarking benchmarks across modalities, we find that, with one exception, none report performance across languages, cultural content types, or population groups. To address this, we propose three concrete evaluation dimensions for pluralistic watermark benchmarking: cross-lingual detection parity, culturally diverse content coverage, and demographic disaggregation of detection metrics. We connect these to the governance frameworks currently mandating watermarking deployment and show that watermarking is held to a lower fairness standard than the generative systems it is meant to govern. Our position is that evaluation must precede deployment, and that the same bias auditing requirements applied to AI models should extend to the verification layer.

Watermarking schemes for large language model output are evaluated almost exclusively on English text using each scheme's detection threshold and a narrow set of quality measurements. Multilingual deployment exposes evaluation-design choices that are inconsequential on English but determine conclusions cross-lingually. We propose an evaluation framework with four components: detection thresholds calibrated empirically per deployment context, a threshold-independent companion measurement that distinguishes calibration failures from detection failures, three disjoint quality measurement paradigms (distributional, paired-semantic, and reference-perplexity), and a generalized-entropy decomposition of cross-language disparity over a typological family partition. Applied to six watermarking schemes, three open-weight generators, eleven languages spanning four scripts and eight typological families, and both base and instruction-tuned regimes, the framework reveals failure modes that single-language single-paradigm evaluation cannot surface. Across detection and quality, observed disparity is predominantly between-family on the typological partition, indicating that cross-lingual fairness gaps in watermarking are structural to language properties rather than idiosyncratic to particular languages.

Domestic robots rely on rich perception to operate in private homes, but privacy risk persists even when raw sensor data remain local. Structured representations exported to downstream planners, cloud services, logs, or learning pipelines can still reveal household information through semantics, geometry, spatial structure, and task targets. We introduce Task-Functional Perception Distillation (TFPD), a task-scoped representation-export framework that keeps rich perception local and profiles downstream exports according to task utility, direct exposure, and multiple residual inference risks. Using 120 AI2-THOR scenes with scene-disjoint train/validation/test splits, frozen attacker selection, and representation-aware held-out attacks, we evaluate navigation, collision checking, and object-goal execution. Three navigation exports achieve identical success (1.000) and mean path ratio (0.898), yet representation-level linkability ranges from 0.532 to 0.970. Replacing an explicit target label with a target region reduces target-category macro-F1 from 1.000 to 0.077 while preserving success at 0.995, while geometric coarsening reduces object-category macro-F1 from 0.704 to 0.556 at a measurable collision-utility cost. A ProcTHOR replication preserves the navigation task-equivalence/privacy-inequivalence finding while changing the relative ordering of normalized and topological exports. These results show that neither field removal nor stronger abstraction induces a universal privacy ordering and motivate task-specific, multi-risk evaluation of the complete public representation.

Membership inference attacks (MIAs) threaten the privacy of machine learning models by revealing whether a specific data point was used during training. Existing MIAs often rely on impractical assumptions such as access to public datasets, shadow models, confidence scores, or training data distribution knowledge and making them vulnerable to defenses like confidence masking and adversarial regularization. Label-only MIAs, even under strict constraints suffer from high query requirements per sample. We propose a cost-effective label-only MIA framework based on transferability and model extraction. By querying the target model M using active sampling, perturbation-based selection, and synthetic data, we extract a functionally similar surrogate S on which membership inference is performed. This shifts query overhead to a one-time extraction phase, eliminating repeated queries to M . Operating under strict black-box constraints, our method matches the performance of state-of-the-art label-only MIAs while significantly reducing query costs. On benchmarks including Purchase, Location, and Texas Hospital, we show that a query budget equivalent to testing ≈1% of training samples suffices to extract S and achieve membership inference accuracy within ±1% of M . We also evaluate the effectiveness of standard defenses proposed for label-only MIAs against our attack.