AI chatbots are increasingly used for answering health-related questions. This study examines the role of topic type discussed with an AI chatbot and individual characteristics on perceived benefits and risks, intention to use an AI chatbot, and willingness to self-disclose health information. We conducted an online experiment with a 2 (topic type: physical versus psychological, between-subjects) x 2 (topic sensitivity: low versus high, within-subjects) mixed design among a Dutch representative sample (N = 1,388). Results showed that perceived benefits were positively associated with intention and willingness to self-disclose, while perceived risks were negatively associated. Moreover, participants reported higher usage intentions for low-sensitive topics compared to high-sensitive topics. Furthermore, perceptions, intention, and willingness to self-disclose varied by individual characteristics. Overall, our findings suggest that intentions to use AI chatbots and self-disclosure of health-related information are primarily related to perceived benefits and risks and to personal characteristics rather than to topic type.
The increasing integration of artificial intelligence (AI) in higher education has raised important questions regarding students' transparency in reporting AI-assisted work. This study investigates the psychological mechanisms underlying university students' willingness to disclose AI use by applying the Cognition--Affect--Conation (CAC) framework. A sequential explanatory mixed-methods design was employed. In the quantitative phase, survey data were collected from 546 university students and analysed using structural equation modelling to examine the relationships among cognitive perceptions, affective responses, and disclosure intention. In the qualitative phase, semi-structured interviews with 22 students were conducted to further interpret the quantitative findings. The results indicate that psychological safety significantly increases students' willingness to disclose AI use and is positively shaped by perceived fairness, perceived teacher support, and perceived organisational support. Conversely, evaluation apprehension reduces disclosure intention and psychological safety, and is strengthened by perceived stigma, perceived uncertainty, and privacy concern. Qualitative findin
The additional information carried by an enlarged filtration and its measurement was studied by several authors. Already Meyer (Sur un theoreme de J. Jacod, 1978) and Yor (Entropie d'une partition, et grossissement initial d'une filtration, 1985), investigated stability of martingale spaces with respect to initial enlargement with atomic sigma-field. We extend these considerations to the case where information is disclosed progressively in time. We define the entropy of such information and we prove that its finiteness is enough for stability of some martingale spaces in progressive setting. Finally we calculate additional logarithmic utility of a discrete information disclosed progressively in time.
Frontier AI developers are increasingly deploying highly capable models internally to automate AI R&D, but these deployments currently face limited external oversight. It is essential, therefore, that developers provide evidence that internally deployed models are safe. While recent work has highlighted the risks of internal deployments and proposed broad approaches to transparency and governance, there remains little guidance on the specific information developers should disclose about them. We address this gap by identifying key information that companies should disclose about internally deployed models across four categories: capabilities, usage, safety mitigations, and governance. For each category, we analyse the key benefits and limitations of disclosure and consider how disclosure-related risks can be mitigated. Our framework could be used by developers to inform both public transparency documents, such as model system cards, and private periodic reports required under emerging frontier AI regulation.
AI systems are increasingly deployed in conversational settings where users may be uncertain whether they are speaking with a human or an AI. Despite mounting regulatory attention to this known safety risk, existing evaluations of AI disclosure are typically English-only, based on machine-generated questions, and restricted to text. We present RealityTest to comprehensively test whether AI systems disclose their identity when asked. The benchmark is the first large-scale multimodal and multilingual evaluation, grounded in human data on how people actually encounter and question AI identity in the real-world. Alongside the benchmark, we release the underlying dataset of 3,152 identity-probing queries collected from ~750 participants across 49 countries and five languages, in text and speech scenarios. We find that only 31% of people ask about identity directly in ambiguous scenarios, and that the questions people ask are far more diverse than machine-generated queries. We test 17 text and 6 speech models, and find substantial variation in disclosure behaviour. However, a single suppression instruction reduces disclosure rates to below 30%, even in the best-performing models. Validat
We read twelve well-known LLM agent benchmark papers and recorded, dimension by dimension, what each paper actually says about how its evaluation was run. The motivation came from a familiar frustration: two papers will report results on the same benchmark with the same model name and disagree, and you cannot tell why -- the scaffold, the sampling settings, the subset, or the evaluator version. In many cases the published artifact does not let you answer. This paper is an implementation report on the attempt. We designed a small audit schema (five fields: benchmark identity, harness specification, inference settings, cost reporting, failure breakdown), wrote a scoring codebook with the boundary cases we hit during pilot scoring, applied it to twelve canonical papers (eight agent, four classical static), and recorded what we saw. We score the disclosure of an agent run, not its correctness, and make no claim that disclosure implies a trustworthy result. The mean audit score across the eight agent-benchmark papers is 0.38 (out of 1.0), and across the four classical static benchmarks 0.66; the largest gap is on cost (none of the eight agent benchmark papers disclose inference cost in
Have you ever read a blog or social media post and suspected that it was written--at least in part--by artificial intelligence (AI)? While transparently acknowledging contributors to writing is generally valued, why some writers choose to disclose or withhold AI involvement remains unclear. In this work, we ask what factors shape writers' decisions to disclose their AI use as a starting point to effectively advocate for transparency. To shed light on this question, we synthesize study findings and theoretical frameworks in human-AI interaction and behavioral science. Concretely, we identify and curate a list of factors that could affect writers' decisions regarding disclosure for human-AI co-created content.
Collecting data on sensitive topics remains challenging in HCI, as participants often withhold information due to privacy concerns and social desirability bias. While chatbots' perceived anonymity may reduce these barriers, research paradoxically suggests people tend to over-share personal or sensitive information with chatbots. In this work, we explore privacy controls in chatbot interviews to address this problem. The privacy control allows participants to revise their transcripts at the end of the interview, featuring two design variants: free editing and AI-aided editing. In a between-subjects study \red{($N=188$)}, we compared no-editing, free-editing, and AI-aided editing conditions in a chatbot-based interview on a sensitive topic. Our results confirm the prevalent issue of oversharing in chatbot-based interviews and show that AI-aided editing serves as an effective privacy-control mechanism, reducing PII disclosure while maintaining data quality and user engagement, thereby offering a promising approach to balancing ethical practice and data quality in such interviews.
Mechanistic interpretability papers increasingly use causal vocabulary: circuits, mediators, causal abstraction, monosemanticity. Such claims require explicit identification assumptions. A purposive audit of 10 papers across four methodological strands finds no dedicated identification-assumptions section and a recurring pattern: validation metrics such as faithfulness, completeness, monosemanticity, alignment, or ablation effects are reported as causal support without stating the assumptions that make them identifying. A two-human-coder audit on $n=30$ reproduces the direction of the main finding: dedicated identification sections are absent, and validation-metric substitution is common, though exact Dim B/D counts are coding-rule sensitive. The paper proposes a disclosure norm: state whether the claim is causal, name the identification strategy, enumerate assumptions, stress at least one, and explain how conclusions shift if assumptions fail. Validation is not identification.
It is commonly accepted that clinicians are ethically obligated to disclose their use of medical machine learning systems to patients, and that failure to do so would amount to a moral fault for which clinicians ought to be held accountable. Call this "the disclosure thesis." Four main arguments have been, or could be, given to support the disclosure thesis in the ethics literature: the risk-based argument, the rights-based argument, the materiality argument, and the autonomy argument. In this article, I argue that each of these four arguments are unconvincing, and therefore, that the disclosure thesis ought to be rejected. I suggest that mandating disclosure may also even risk harming patients by providing stakeholders with a way to avoid accountability for harm that results from improper applications or uses of these systems.
AI is rapidly transforming journalism, but the extent of its use in published newspaper articles remains unclear. We address this gap by auditing a large-scale dataset of 186K articles from online editions of 1.5K American newspapers published in the summer of 2025. Using Pangram, a state-of-the-art AI detector, we discover that approximately 9% of newly-published articles are either partially or fully AI-generated. This AI use is unevenly distributed, appearing more frequently in smaller, local outlets, in specific topics such as weather and technology, and within certain ownership groups. We also analyze 45K opinion pieces from Washington Post, New York Times, and Wall Street Journal, finding that they are 6.4 times more likely to contain AI-generated content than news articles from the same publications, with many AI-flagged op-eds authored by prominent public figures. Despite this prevalence, we find that AI use is rarely disclosed: a manual audit of 100 AI-flagged articles found only five disclosures of AI use. Overall, our audit highlights the immediate need for greater transparency and updated editorial standards regarding the use of AI in journalism to maintain public trust
Large language models are increasingly deployed in multi-agent settings whose outcomes hinge on social intelligence, motivating evaluations of their interactive capabilities; yet existing studies remain overwhelmingly empirical, leaving us without a theoretical understanding of how agent interactions determine collective outcomes. To address this, we introduce \textit{Mini-Mafia}, a four-player simplification of the social deduction game Mafia in which a fixed night phase reduces the game to a single critical exchange among a mafioso, a detective, and a villager. In this setting, we show that the mafia win-rate $p$ is predicted by the analytical formula $\text{logit}(p) = v \times (m - d)$, where $m$, $d$, and $v$ represent the mafioso's deception, the detective's disclosure, and the villager's detection capabilities. We turn this analytical framework into the \textit{Mini-Mafia Benchmark}, where Bayesian inference over gameplay data yields per-model estimates of the intrinsic parameters $m$, $d$, and $v$. For $I$ models, only $3I$ parameters suffice to predict the outcomes of all $I^3$ tournament combinations; and in 5-fold cross-validation the formula achieves a $76.6\%$ Brier-sc
Although encrypted control systems ensure confidentiality of private data, it is challenging to detect anomalies without the secret key as all signals remain encrypted. To address this issue, we propose a homomorphic encryption scheme for dynamic controllers that automatically discloses the residue signal for anomaly detection, while keeping all other signals private. To this end, we characterize the zero-dynamics of an encrypted dynamic system over a finite field of integers and incorporate it into a Learning With Errors (LWE) based scheme. We then present a method to further utilize the disclosed residue signal for implementing dynamic controllers over encrypted data, which does not involve re-encryption even when they have non-integer state matrices.
Self-disclosed mental health diagnoses, which serve as ground truth annotations of mental health status in the absence of clinical measures, underpin the conclusions behind most computational studies of mental health language from the last decade. However, psychiatric conditions are dynamic; a prior depression diagnosis may no longer be indicative of an individual's mental health, either due to treatment or other mitigating factors. We ask: to what extent are self-disclosures of mental health diagnoses actually relevant over time? We analyze recent activity from individuals who disclosed a depression diagnosis on social media over five years ago and, in turn, acquire a new understanding of how presentations of mental health status on social media manifest longitudinally. We also provide expanded evidence for the presence of personality-related biases in datasets curated using self-disclosed diagnoses. Our findings motivate three practical recommendations for improving mental health datasets curated using self-disclosed diagnoses: 1) Annotate diagnosis dates and psychiatric comorbidities; 2) Sample control groups using propensity score matching; 3) Identify and remove spurious corre
Strategic information disclosure, in its simplest form, considers a game between an information provider (sender) who has access to some private information that an information receiver is interested in. While the receiver takes an action that affects the utilities of both players, the sender can design information (or modify beliefs) of the receiver through signal commitment, hence posing a Stackelberg game. However, obtaining a Stackelberg equilibrium for this game traditionally requires the sender to have access to the receiver's objective. In this work, we consider an online version of information design where a sender interacts with a receiver of an unknown type who is adversarially chosen at each round. Restricting attention to Gaussian prior and quadratic costs for the sender and the receiver, we show that $\mathcal{O}(\sqrt{T})$ regret is achievable with full information feedback, where $T$ is the total number of interactions between the sender and the receiver. Further, we propose a novel parametrization that allows the sender to achieve $\mathcal{O}(\sqrt{T})$ regret for a general convex utility function. We then consider the Bayesian Persuasion problem with an additional
We consider the disclosure problem of a sender with a large data set of hard evidence who wants to persuade a receiver to take higher actions. Because the receiver will make inferences based on the distribution of the data they see, the sender has an incentive to drop observations to mimic the distributions that would be observed under better states. We predict which observations the sender discloses using a model that approximates large datasets with a continuum of data. It is optimal for the sender to play an imitation strategy, under which they submit evidence that imitates the natural distribution under some more desirable target state. We characterize the partial-pooling outcomes under these imitation strategies, and show that they are supported by data on the outcomes that maximally distinguish higher states. Relative to full information, the equilibrium with voluntary disclosure reduces the welfare of senders with little data or a favorable state, who fully disclose their data but suffer the receiver's skepticism, and benefits senders with access to large datasets, who can profitably drop observations under low states.
A firm that discloses data about one customer moves a downstream seller's belief about every correlated customer, pricing third parties it never transacts with. This privacy externality equals the change in downstream deadweight loss, is signed by which side of the pricing threshold a customer is on, and falls hardest on those just carried across. Disclosure is privately optimal on an open set of imperfect correlations; incentive compatibility prices it through a distorted allocation and rations the discount at the top under a continuum of types. Selling tips disclosure past a liquidity threshold; consent dominates both data minimization and laissez-faire.
The increasing adoption of differential privacy (DP) leads to public-facing DP deployments by both government agencies and companies. However, real-world DP deployments often do not fully disclose their privacy guarantees, which vary greatly between deployments. Failure to disclose certain DP parameters can lead to misunderstandings about the strength of the privacy guarantee, undermining the trust in DP. In this work, we seek to inform future standards for communicating the privacy guarantees of DP deployments. Based on semi-structured interviews with 12 DP experts, we identify important DP parameters necessary to comprehensively communicate DP guarantees, and describe why and how they should be disclosed. Based on expert recommendations, we design an initial privacy label for DP to comprehensively communicate privacy guarantees in a standardized format.
We propose a model of causal persuasion, in which a sender selectively discloses a set of variables together with their true joint distribution and proposes a subjective causal model that binds them. A receiver is persuaded by this model only if the data conclusively identifies the causal link of interest. We characterize when such persuasion succeeds or fails, and how easily it can be achieved. We further show that if the receiver holds a pre-existing subjective model, debunking it is similar to persuading a receiver without one. To establish a true causal link, the sender often needs to disclose only one or two well-chosen variables. But to dispel a perceived link -- to persuade the receiver there is no causal relationship -- every common cause must be disclosed. Our results highlight a fundamental asymmetry in causal persuasion: Establishing causality is often much easier than ruling it out.
In many decision-making scenarios, individuals strategically choose what information to disclose to optimize their own outcomes. It is unclear whether such strategic information disclosure can lead to good societal outcomes. To address this question, we consider a competitive Bayesian persuasion model in which multiple agents selectively disclose information about their qualities to a principal, who aims to choose the candidates with the highest qualities. Using the price-of-anarchy framework, we quantify the inefficiency of such strategic disclosure. We show that the price of anarchy is at most a constant when the agents have independent quality distributions, even if their utility functions are heterogeneous. This result provides the first theoretical guarantee on the limits of inefficiency in Bayesian persuasion with competitive information disclosure.