arXiv subjects

Carson Ezell

Publications and source records attributed to Carson Ezell.

15 recordsLinked to original sources

arXiv2026

FLARE-AI: Flaw Reporting for AI

Flaw reporting for deployed AI systems is fundamental to identifying system failures and improving AI safety. Yet the AI reporting ecosystem is fragmented: researchers who identify flaws often do not know what or where to report, and groups who receive reports rarely share them with other relevant stakeholders. As a result, good-faith reporters duplicate effort by submitting many different forms, and recipients lack standardized, triage-ready information. We audit 12 reporting systems published by AI developers, cybersecurity groups, and AI flaw aggregators, identifying five recurring design challenges spanning discoverability, scope, information collection, coordination, and guidance for strict-liability cases. Building on this analysis and feedback from 49 experts across 32 organizations representing developers, security researchers, and ecosystem coordinators, we introduce FLARE-AI, an open-source AI flaw reporting system designed for interoperability with existing systems. FLARE-AI streamlines flaw report creation by collecting triage-relevant information through conditional logic and early classification, then enables optional dissemination of standardized, machine-readable reports to multiple developers, coordinators, and incident registries from a single submission. By lowering barriers to reporting AI flaws and improving interoperability across stakeholders, FLARE-AI helps break down silos and accelerate remediation across the AI ecosystem.

cs.CY↗

arXiv2026

RCTs for Frontier AI Governance: Methodological Challenges and Solutions for Human Uplift Studies

Human uplift studies, or studies that measure the effects of AI access on human performance via randomized controlled trials (RCT) or similar methodologies, increasingly inform frontier AI governance and deployment decisions. While RCT methods are robust in other fields, their interaction with the distinctive properties of frontier AI systems remains underexamined, particularly when results are used to inform high-stakes decisions. We present findings from interviews with 16 expert practitioners with experience conducting human uplift studies in domains including biosecurity, cybersecurity, education, and labor. Across interviews, experts described a recurring tension between the standard causal inference assumptions upon which human uplift studies rely and the object of study itself. Rapidly evolving AI systems, shifting baselines, heterogeneous and changing user proficiency, and porous real-world settings strain assumptions underlying internal, external, and construct validity, complicating the interpretation and appropriate use of uplift evidence. We contribute (1) a synthesis of methodological challenges in human uplift studies, mapped to risks to study validity and classified by their degree of specificity to large language model (LLM) systems, and (2) a mapping from challenges to proposed solutions. By collating expert-identified challenges and solutions, we seek to clarify the interpretive limits and appropriate uses of human uplift evidence, to align evaluation practice with the decisions it informs, and to support more coordinated methodological foundations for AI governance.

cs.CY↗

arXiv2025

Dark Speculation: Combining Qualitative and Quantitative Understanding in Frontier AI Risk Analysis

Estimating catastrophic harms from frontier AI is hindered by deep ambiguity: many of its risks are not only unobserved but unanticipated by analysts. The central limitation of current risk analysis is the inability to populate the $\textit{catastrophic event space}$, or the set of potential large-scale harms to which probabilities might be assigned. This intractability is worsened by the $\textit{Lucretius problem}$, or the tendency to infer future risks only from past experience. We propose a process of $\textit{dark speculation}$, in which systematically generating and refining catastrophic scenarios ("qualitative" work) is coupled with estimating their likelihoods and associated damages (quantitative underwriting analysis). The idea is neither to predict the future nor to enable insurance for its own sake, but to use narrative and underwriting tools together to generate probability distributions over outcomes. We formalize this process using a simplified catastrophic L\'{e}vy stochastic framework and propose an iterative institutional design in which (1) speculation (including scenario planning) generates detailed catastrophic event narratives, (2) insurance underwriters assign probabilistic and financial parameters to these narratives, and (3) decision-makers synthesize the results into summary statistics to inform judgment. Analysis of the model reveals the value of (a) maintaining independence between speculation and underwriting, (b) analyzing multiple risk categories in parallel, and (c) generating "thick" catastrophic narrative rich in causal (counterfactual) and mitigative detail. While the approach cannot eliminate deep ambiguity, it offers a systematic approach to reason about extreme, low-probability events in frontier AI, tempering complacency and overreaction. The framework is adaptable for iterative use and can be further augmented with AI systems.

cs.CY↗

arXiv2025

Incident Analysis for AI Agents

As AI agents become more widely deployed, we are likely to see an increasing number of incidents: events involving AI agent use that directly or indirectly cause harm. For example, agents could be prompt-injected to exfiltrate private information or make unauthorized purchases. Structured information about such incidents (e.g., user prompts) can help us understand their causes and prevent future occurrences. However, existing incident reporting processes are not sufficient for understanding agent incidents. In particular, such processes are largely based on publicly available data, which excludes useful, but potentially sensitive, information such as an agent's chain of thought or browser history. To inform the development of new, emerging incident reporting processes, we propose an incident analysis framework for agents. Drawing on systems safety approaches, our framework proposes three types of factors that can cause incidents: system-related (e.g., CBRN training data), contextual (e.g., prompt injections), and cognitive (e.g., misunderstanding a user request). We also identify specific information that could help clarify which factors are relevant to a given incident: activity logs, system documentation and access, and information about the tools an agent uses. We provide recommendations for 1) what information incident reports should include and 2) what information developers and deployers should retain and make available to incident investigators upon request. As we transition to a world with more agents, understanding agent incidents will become increasingly crucial for managing risks.

cs.CY↗

arXiv2025

Multi-Agent Risks from Advanced AI

The rapid development of advanced AI agents and the imminent deployment of many instances of these agents will give rise to multi-agent systems of unprecedented complexity. These systems pose novel and under-explored risks. In this report, we provide a structured taxonomy of these risks by identifying three key failure modes (miscoordination, conflict, and collusion) based on agents' incentives, as well as seven key risk factors (information asymmetries, network effects, selection pressures, destabilising dynamics, commitment problems, emergent agency, and multi-agent security) that can underpin them. We highlight several important instances of each risk, as well as promising directions to help mitigate them. By anchoring our analysis in a range of real-world examples and experimental evidence, we illustrate the distinct challenges posed by multi-agent systems and their implications for the safety, governance, and ethics of advanced AI.

cs.MA↗

arXiv2025

The AI Agent Index

Leading AI developers and startups are increasingly deploying agentic AI systems that can plan and execute complex tasks with limited human involvement. However, there is currently no structured framework for documenting the technical components, intended uses, and safety features of agentic systems. To fill this gap, we introduce the AI Agent Index, the first public database to document information about currently deployed agentic AI systems. For each system that meets the criteria for inclusion in the index, we document the system's components (e.g., base model, reasoning implementation, tool use), application domains (e.g., computer use, software engineering), and risk management practices (e.g., evaluation results, guardrails), based on publicly available information and correspondence with developers. We find that while developers generally provide ample information regarding the capabilities and applications of agentic systems, they currently provide limited information regarding safety and risk management practices. The AI Agent Index is available online at https://aiagentindex.mit.edu/

cs.SE↗

arXiv2024

How Do AI Companies "Fine-Tune" Policy? Examining Regulatory Capture in AI Governance

Industry actors in the United States have gained extensive influence in conversations about the regulation of general-purpose artificial intelligence (AI) systems. Although industry participation is an important part of the policy process, it can also cause regulatory capture, whereby industry co-opts regulatory regimes to prioritize private over public welfare. Capture of AI policy by AI developers and deployers could hinder such regulatory goals as ensuring the safety, fairness, beneficence, transparency, or innovation of general-purpose AI systems. In this paper, we first introduce different models of regulatory capture from the social science literature. We then present results from interviews with 17 AI policy experts on what policy outcomes could compose regulatory capture in US AI policy, which AI industry actors are influencing the policy process, and whether and how AI industry actors attempt to achieve outcomes of regulatory capture. Experts were primarily concerned with capture leading to a lack of AI regulation, weak regulation, or regulation that over-emphasizes certain policy goals over others. Experts most commonly identified agenda-setting (15 of 17 interviews), advocacy (13), academic capture (10), information management (9), cultural capture through status (7), and media capture (7) as channels for industry influence. To mitigate these particular forms of industry influence, we recommend systemic changes in developing technical expertise in government and civil society, independent funding streams for the AI ecosystem, increased transparency and ethics requirements, greater civil society access to policy, and various procedural safeguards.

cs.CY↗

arXiv2024

An FDA for AI? Pitfalls and Plausibility of Approval Regulation for Frontier Artificial Intelligence

Observers and practitioners of artificial intelligence (AI) have proposed an FDA-style licensing regime for the most advanced AI models, or 'frontier' models. In this paper, we explore the applicability of approval regulation -- that is, regulation of a product that combines experimental minima with government licensure conditioned partially or fully upon that experimentation -- to the regulation of frontier AI. There are a number of reasons to believe that approval regulation, simplistically applied, would be inapposite for frontier AI risks. Domains of weak fit include the difficulty of defining the regulated product, the presence of Knightian uncertainty or deep ambiguity about harms from AI, the potentially transmissible nature of risks, and distributed activities among actors involved in the AI lifecycle. We conclude by highlighting the role of policy learning and experimentation in regulatory development, describing how learning from other forms of AI regulation and improvements in evaluation and testing methods can help to overcome some of the challenges we identify.

cs.CY↗

arXiv2024

Black-Box Access is Insufficient for Rigorous AI Audits

External audits of AI systems are increasingly recognized as a key mechanism for AI governance. The effectiveness of an audit, however, depends on the degree of access granted to auditors. Recent audits of state-of-the-art AI systems have primarily relied on black-box access, in which auditors can only query the system and observe its outputs. However, white-box access to the system's inner workings (e.g., weights, activations, gradients) allows an auditor to perform stronger attacks, more thoroughly interpret models, and conduct fine-tuning. Meanwhile, outside-the-box access to training and deployment information (e.g., methodology, code, documentation, data, deployment details, findings from internal evaluations) allows auditors to scrutinize the development process and design more targeted evaluations. In this paper, we examine the limitations of black-box audits and the advantages of white- and outside-the-box audits. We also discuss technical, physical, and legal safeguards for performing these audits with minimal security risks. Given that different forms of access can lead to very different levels of evaluation, we conclude that (1) transparency regarding the access and methods used by auditors is necessary to properly interpret audit results, and (2) white- and outside-the-box access allow for substantially more scrutiny than black-box access alone.

cs.CY↗

arXiv2024

Visibility into AI Agents

Increased delegation of commercial, scientific, governmental, and personal activities to AI agents -- systems capable of pursuing complex goals with limited supervision -- may exacerbate existing societal risks and introduce new risks. Understanding and mitigating these risks involves critically evaluating existing governance structures, revising and adapting these structures where needed, and ensuring accountability of key stakeholders. Information about where, why, how, and by whom certain AI agents are used, which we refer to as visibility, is critical to these objectives. In this paper, we assess three categories of measures to increase visibility into AI agents: agent identifiers, real-time monitoring, and activity logging. For each, we outline potential implementations that vary in intrusiveness and informativeness. We analyze how the measures apply across a spectrum of centralized through decentralized deployment contexts, accounting for various actors in the supply chain including hardware and software service providers. Finally, we discuss the implications of our measures for privacy and concentration of power. Further work into understanding the measures and mitigating their negative impacts can help to build a foundation for the governance of AI agents.

cs.CY↗

arXiv2023

SkyWatch: A Passive Multistatic Radar Network for the Measurement of Object Position and Velocity

(Abridged) Quantitative three-dimensional (3D) position and velocity estimates obtained by passive radar will assist the Galileo Project in the detection and classification of aerial objects by providing critical measurements of range, location, and kinematics. These parameters will be combined with those derived from the Project{\textquoteright}s suite of electromagnetic sensors and used to separate known aerial objects from those exhibiting anomalous kinematics. SkyWatch, a passive multistatic radar system based on commercial broadcast FM radio transmitters of opportunity, is a network of receivers spaced at geographical scales that enables estimation of the 3D position and velocity time series of objects at altitudes up to 80km, horizontal distances up to 150km, and at velocities to {\textpm}2{\textpm}2km/s ({\textpm}6{\textpm}6Mach). The receivers are designed to collect useful data in a variety of environments varying by terrain, transmitter power, relative transmitter distance, adjacent channel strength, etc. In some cases, the direct signal from the transmitter may be large enough to be used as the reference with which the echoes are correlated. In other cases, the direct signal may be weak or absent, in which case a reference is communicated to the receiver from another network node via the internet for echo correlation. Various techniques are discussed specific to the two modes of operation and a hybrid mode. Delay and Doppler data are sent via internet to a central server where triangulation is used to deduce time series of 3D positions and velocities. A multiple receiver (multistatic) radar experiment is undergoing Phase 1 testing, with several receivers placed at various distances around the Harvard{\textendash}Smithsonian Center for Astrophysics (CfA), to validate full 3D position and velocity recovery.

astro-ph.IM↗

arXiv2023

The Scientific Investigation of Unidentified Aerial Phenomena (UAP) Using Multimodal Ground-Based Observatories

(Abridged) Unidentified Aerial Phenomena (UAP) have resisted explanation and have received little formal scientific attention for 75 years. A primary objective of the Galileo Project is to build an integrated software and instrumentation system designed to conduct a multimodal census of aerial phenomena and to recognize anomalies. Here we present key motivations for the study of UAP and address historical objections to this research. We describe an approach for highlighting outlier events in the high-dimensional parameter space of our census measurements. We provide a detailed roadmap for deciding measurement requirements, as well as a science traceability matrix (STM) for connecting sought-after physical parameters to observables and instrument requirements. We also discuss potential strategies for deciding where to locate instruments for development, testing, and final deployment. Our instrument package is multimodal and multispectral, consisting of (1) wide-field cameras in multiple bands for targeting and tracking of aerial objects and deriving their positions and kinematics using triangulation; (2) narrow-field instruments including cameras for characterizing morphology, spectra, polarimetry, and photometry; (3) passive multistatic arrays of antennas and receivers for radar-derived range and kinematics; (4) radio spectrum analyzers to measure radio and microwave emissions; (5) microphones for sampling acoustic emissions in the infrasonic through ultrasonic frequency bands; and (6) environmental sensors for characterizing ambient conditions (temperature, pressure, humidity, and wind velocity), as well as quasistatic electric and magnetic fields, and energetic particles. The use of multispectral instruments and multiple sensor modalities will help to ensure that artifacts are recognized and that true detections are corroborated and verifiable.

astro-ph.IM↗

arXiv2023

Detection Rate of <50-meter Interstellar Objects with LSST

The previous decade saw the discovery of the first four known interstellar objects due to advances in astronomical viewing equipment. Future sky surveys with greater sensitivity will allow for more frequent detections of such objects, including increasingly small objects. We consider the capabilities of the Legacy Survey of Space and Time (LSST) of the Vera C. Rubin Observatory to detect interstellar objects of small sizes during its period of operation over the next decade. We use LSST's detection capabilities and simulate populations of interstellar objects in the range of 1-50m in diameter to calculate the expected number of small interstellar objects that will be detected. We use previous detections of interstellar objects to calibrate our object density estimates. We also consider the impact of the population's albedo on detection rates by considering populations with two separate albedo distributions: a constant albedo of 0.06 and an albedo distribution that resembles near earth asteroids. We find that the number of detections increases with the diameter over the range of diameters we consider. We estimate a detection rate of up to a small ISO every two years of LSST's operation with an increase by a factor of ten for future surveys that extend a magnitude deeper.

astro-ph.IM↗

arXiv2022

A Lunar Backup Record of Humanity

The risk of a catastrophic or existential disaster for our civilization is increasing this century. A significant motivation for a near-term space settlement is the opportunity to safeguard civilization in the event of a planetary-scale disaster. A catastrophic event could destroy the significant cultural, scientific, and technological progress on Earth. However, early space settlements can preserve records of human activity by maintaining a backup data storage system. The backup can also store information about the events leading up to the disaster. The system would improve the ability of early space settlers to recover our civilization after collapse. We show that advances in laser communications and data storage enable the development of a data storage system on the lunar surface with a sufficient uplink data rate and storage capacity to preserve valuable information about the achievements of our civilization and the chronology of the disaster.

physics.soc-ph↗

arXiv2022

The Inferred Abundance of Interstellar Objects of Technological Origin

The local detection rate of interstellar objects can allow for estimations of the total number of similar objects bound by the Milky Way thin disk. If interstellar objects of artificial origin are discovered, the estimated total number of objects can be lower by a factor of about $10^{16}$ if they target the habitable zone around the Sun. We propose a model for calculating the quantity of natural or artificial interstellar objects of interest based on the object's velocity and observed density. We then apply the model to the case of chemically propelled rockets from extraterrestrial civilizations. Finally, we apply the model to three previously discovered interstellar objects -- the object 'Oumuamua of unknown origin and the first interstellar meteors CNEOS 2014-01-08 and CNEOS 2017-03-09.

physics.pop-ph↗