arXiv subjects

Himanshu Joshi

Publications and source records attributed to Himanshu Joshi.

16 recordsLinked to original sources

arXiv2026

Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts

Artificial intelligence poses many risks, ranging from familiar present-day harms to unprecedented and potentially catastrophic ones. Effective risk management requires prioritization: we must understand which risks are most severe, who is most vulnerable, and who is most responsible for addressing them. We report results from a three-round Delphi study conducted late 2025 with 272 international AI experts. Experts rated 24 AI risks on harm probability and severity, sector and actor vulnerability, actor responsibility, and overall concern. Experts estimated the five most severe harms in the next 5 years were likely to come from dangerous capabilities, competitive dynamics, weapons & cyberattacks (including CBRNE), power centralization, and false information. In a business-as-usual scenario, experts judged 18 of 24 risks as having a more than 10% probability of catastrophic outcomes (e.g., more than 1 million deaths or more than USD 100B in financial loss) in the next 5 years (2025-2030). In a scenario where pragmatic mitigations are implemented, experts still judged five risks as having a more than 10% probability of catastrophic outcomes: dangerous capabilities, weapons & cyberattacks, environmental harm, inequality & unemployment, and power centralization. All 24 risks were judged as being more than 5% likely to cause catastrophic outcomes. AI users and the general public were judged the most vulnerable to these risks, but experts assigned the highest responsibility for addressing them to general-purpose AI developers and governance actors (including governments, regulators, and standards bodies). Across most risks, experts identified information, finance, and national security as the most vulnerable sectors. These findings can guide AI risk prioritization and clarify expert expectations about who should bear responsibility for mitigation.

cs.CY↗

arXiv2026

Interpreting Agentic Systems: Beyond Model Explanations to System-Level Accountability

Agentic systems have transformed how Large Language Models (LLMs) can be leveraged to create autonomous systems with goal-directed behaviors, consisting of multi-step planning and the ability to interact with different environments. These systems differ fundamentally from traditional machine learning models, both in architecture and deployment, introducing unique AI safety challenges, including goal misalignment, compounding decision errors, and coordination risks among interacting agents, that necessitate embedding interpretability and explainability by design to ensure traceability and accountability across their autonomous behaviors. Current interpretability techniques, developed primarily for static models, show limitations when applied to agentic systems. The temporal dynamics, compounding decisions, and context-dependent behaviors of agentic systems demand new analytical approaches. This paper assesses the suitability and limitations of existing interpretability methods in the context of agentic systems, identifying gaps in their capacity to provide meaningful insight into agent decision-making. We propose future directions for developing interpretability techniques specifically designed for agentic systems, pinpointing where interpretability is required to embed oversight mechanisms across the agent lifecycle from goal formation, through environmental interaction, to outcome evaluation. These advances are essential to ensure the safe and accountable deployment of agentic AI systems.

cs.AI↗

arXiv2025

A Stochastic Differential Equation Framework for Multi-Objective LLM Interactions: Dynamical Systems Analysis with Code Generation Applications

We introduce a general stochastic differential equation framework for modelling multiobjective optimization dynamics in iterative Large Language Model (LLM) interactions. Our framework captures the inherent stochasticity of LLM responses through explicit diffusion terms and reveals systematic interference patterns between competing objectives via an interference matrix formulation. We validate our theoretical framework using iterative code generation as a proof-of-concept application, analyzing 400 sessions across security, efficiency, and functionality objectives. Our results demonstrate strategy-dependent convergence behaviors with rates ranging from 0.33 to 1.29, and predictive accuracy achieving R2 = 0.74 for balanced approaches. This work proposes the feasibility of dynamical systems analysis for multi-objective LLM interactions, with code generation serving as an initial validation domain.

cs.LG↗

arXiv2025

Security Degradation in Iterative AI Code Generation -- A Systematic Analysis of the Paradox

The rapid adoption of Large Language Models(LLMs) for code generation has transformed software development, yet little attention has been given to how security vulnerabilities evolve through iterative LLM feedback. This paper analyzes security degradation in AI-generated code through a controlled experiment with 400 code samples across 40 rounds of "improvements" using four distinct prompting strategies. Our findings show a 37.6% increase in critical vulnerabilities after just five iterations, with distinct vulnerability patterns emerging across different prompting approaches. This evidence challenges the assumption that iterative LLM refinement improves code security and highlights the essential role of human expertise in the loop. We propose practical guidelines for developers to mitigate these risks, emphasizing the need for robust human validation between LLM iterations to prevent the paradoxical introduction of new security issues during supposedly beneficial code "improvements".

cs.SE↗

arXiv2025

Approaches to Responsible Governance of GenAI in Organizations

PEER-REVIEWED AND ACCEPTED IN IEEE- ISTAS 2025 The rapid evolution of Generative AI (GenAI) has introduced unprecedented opportunities while presenting complex challenges around ethics, accountability, and societal impact. This paper draws on a literature review, established governance frameworks, and industry roundtable discussions to identify core principles for integrating responsible GenAI governance into diverse organizational structures. Our objective is to provide actionable recommendations for a balanced, risk-based governance approach that enables both innovation and oversight. Findings emphasize the need for adaptable risk assessment tools, continuous monitoring practices, and cross-sector collaboration to establish trustworthy GenAI. These insights provide a structured foundation and Responsible GenAI Guide (ResAI) for organizations to align GenAI initiatives with ethical, legal, and operational best practices.

cs.CY↗

arXiv2025

Transparent AI: The Case for Interpretability and Explainability

As artificial intelligence systems increasingly inform high-stakes decisions across sectors, transparency has become foundational to responsible and trustworthy AI implementation. Leveraging our role as a leading institute in advancing AI research and enabling industry adoption, we present key insights and lessons learned from practical interpretability applications across diverse domains. This paper offers actionable strategies and implementation guidance tailored to organizations at varying stages of AI maturity, emphasizing the integration of interpretability as a core design principle rather than a retrospective add-on.

cs.LG↗

arXiv2025

Half-metallicity and anomalous Slater-Pauling behaviour in half-Heusler CrMnSb

This study provides a first-principles insight into half-Heusler CrMnSb to understand its deviation from the conventional Slater-Pauling semiconducting behavior. CrMnSb, having a valence electron count of 18, has been proposed to exhibit compensated ferrimagnetic character instead of the expected nonmagnetic semiconducting ground state. As half-Heusler systems with a valence electron count of 18 are not known to exhibit magnetic ordering, we have investigated the electronic and magnetic properties of CrMnSb using a combination of density functional theory and Green's function-based multiple-scattering theory. We show that, despite satisfying the 18 valence electron Slater-Pauling rule, CrMnSb does not exhibit ground-state nonmagnetic semiconducting behavior. Instead, it reveals a half-metallic, fully compensated ferrimagnetic ground state. This anomaly originates from the presence of localized sublattice moments, resulting from antiparallel alignment between Cr and Mn sublattices, which enforces half-metallic ferrimagnetism despite its ideal 18 valence electron count.

cond-mat.mtrl-sci↗

arXiv2025

The structure and migration of twin boundaries in tetragonal $β$-Sn: an application of machine learning based interatomic potentials

Although atomistic simulations have contributed significantly to our understanding of twin boundary structure and migration in metals and alloys with hexagonal close packed (HCP) crystal structures, few direct atomistic studies of twinning have been conducted for other types of low symmetry materials, in large part due to a lack of reliable interatomic potentials. In this work, we examine twin boundary structure and migration in a tetragonal material, $β$-Sn, comparing high resolution Transmission Electron Microscopy (TEM) images of deformation twins in $β$-Sn to the results of direct atomistic simulations using multiple interatomic potentials. ML-based potentials developed in this work are found to give results consistent with our experimental data, revealing faceted twin boundary structures formed by the nucleation and motion of twinning disconnections. We use bicrystallographic methods in combination with atomistic simulations to analyze the structure, energy and shear coupled migration of observed twin facets in $β$-Sn. In analogy to Prismatic-Basal (PB/BP) interfaces in HCP metals, we discover low energy asymmetric Prismatic-A-plane (PA/AP) interfaces important to twin growth in $β$-Sn. A Moment Tensor Potential (MTP) and Rapid Artificial Neural Network (RANN) interatomic potential suitable for studying twinning and phase transformations in Sn are made publicly available as part of this work.

cond-mat.mtrl-sci↗

arXiv2025

Energetics of the nucleation and glide of disconnection modes in symmetric tilt grain boundaries

GBs evolve by the nucleation and glide of disconnections, which are dislocations with a step character. In this work, motivated by recent success in predicting GB properties such as the shear coupling factor and mobility from the intrinsic properties of disconnections, we develop a systematic method to calculate the energy barriers for the nucleation and glide of individual disconnection modes under arbitrary driving forces. This method combines tools from bicrystallography to enumerate disconnection modes and the NEB method to calculate their energetics, yielding minimum energy paths and atomistic mechanisms for the nucleation and glide of each disconnection mode. We apply the method to accurately predict shear coupling factors of [001] symmetric tilt grain boundaries in Cu. Particular attention is paid to the boundaries where the classical disconnection nucleation model produces incorrect nucleation barriers. We demonstrate that the method can accurately compute the energy barriers and predict the shear coupling factors for low temperature regime.

cond-mat.mtrl-sci↗

arXiv2024

Phase Transition and Magnetic Suppression in Heusler Alloy IrMnAl

This study conducts a comprehensive first-principles investigation of the IrMnAl Heusler alloy, highlighting its magnetic properties and assessing the effects of pressure-induced phase shift. The limitations of Generalized Gradient Approximation in accurately representing the compounds magnetic behavior is addressed by employing the GGA+U approach, which more effectively captures the electronic features. At 5.6 GPa pressure, we observe a novel structural phase transition from the cubic F4-3m to the P4-3m space group. The calculated Curie temperature, combined with the analysis of Mn-Mn exchange interaction, sheds light on the weak ferromagnetism observed in the new phase of the compound.

cond-mat.mtrl-sci↗

arXiv2023

Temperature-induced suppression of structural disproportionation in paramagnetic quantum materials

With the development of electronic structure theory, a new class of materials - quantum ones has been recognized by the community. Traditionally, it has been believed that the properties of such compounds cannot be described within the framework of modern density functional theory, and indeed, more advanced post-mean-field theory methods are needed. Motivated by this, herein, we develop a fundamental understanding of such complex materials using the example of paramagnetic YNiO3, which is experimentally known to exhibit metal-to-insulator phase transition. We show that this material has a temperature-dependent distribution of local structural and spin motifs. Thus, while at low temperatures, YNiO3 has distinct structural disproportionation with the formation of large and small octahedra, as the temperature increases, this disproportionation is suppressed. We also explain the paramagnetic monoclinic to paramagnetic orthorhombic phase transition within the double-well to single-well energy profile, predicting the variation of the corresponding energy profile as a function of octahedral size distribution. In this way, we demonstrate a fundamental understanding of structural phase transitions in quantum materials, giving insight into how it can be used for different applications and what minimum level of theory is needed to describe such types of complex materials correctly.

cond-mat.mtrl-sci↗

arXiv2023

Estimating the causal effects of multiple intermittent treatments with application to COVID-19

To draw real-world evidence about the comparative effectiveness of multiple time-varying treatments on patient survival, we develop a joint marginal structural survival model and a novel weighting strategy to account for time-varying confounding and censoring. Our methods formulate complex longitudinal treatments with multiple start/stop switches as the recurrent events with discontinuous intervals of treatment eligibility. We derive the weights in continuous time to handle a complex longitudinal dataset without the need to discretize or artificially align the measurement times. We further use machine learning models designed for censored survival data with time-varying covariates and the kernel function estimator of the baseline intensity to efficiently estimate the continuous-time weights. Our simulations demonstrate that the proposed methods provide better bias reduction and nominal coverage probability when analyzing observational longitudinal survival data with irregularly spaced time intervals, compared to conventional methods that require aligned measurement time points. We apply the proposed methods to a large-scale COVID-19 dataset to estimate the causal effects of several COVID-19 treatments on the composite of in-hospital mortality and ICU admission.

stat.ME↗

arXiv2017

Structure and Electrical Properties of DNA Nanotubes Embedded in Lipid Bilayer Membranes

Engineering the synthetic nanopores through lipid bilayer membrane to access the interior of a cell is a long persisting challenge in biotechnology. Here, we demonstrate the stability and dynamics of a tile-based 6-helix DNA nanotube (DNT) embedded in POPC lipid bilayer using the analysis of 0.2 microsecond long equilibrium MD simulation trajectories. We observe that the head groups of the lipid molecules close to the lumen cooperatively tilt towards the hydrophilic sugar-phosphate backbone of DNA and form a toroidal structure around the patch of DNT protruding in the membrane. Further, we explore the effect of ionic concentrations to the in-solution structure and stability of the lipid-DNT complex. Transmembrane ionic current measurements for the constant electric field MD simulation provide the I-V characteristics of the water filled DNT lumen in lipid membrane. With increasing salt concentrations, the measured values of transmembrane ionic conductance of the porous DNT lumen vary from 4.3 nS to 20.6 nS. Simulations of the DNTs with ssDNA and dsDNA overhangs at the mouth of the pore show gating effect with remarkable difference in the transmembrane ionic conductivities for open and close state nanopores.

cond-mat.soft↗

arXiv2017

Probing the Structure and in Silico Stability of Cargo Loaded DNA Icosahedron using MD Simulations

Platonic solids such as polyhedra based on DNA have been deployed for multifarious applications such as RNAi delivery, biological targeting and bioimaging. All of these applications hinge on the capability of DNA polyhedra for molecular display with high spatial precision. Therefore high resolution structural models of such polyhedra are critical to widen their applications in both materials and biology. Here, we present an atomistic model of a well-characterized DNA icosahedron, with demonstrated versatile functionalities in biological systems. We study the structure and dynamics of this DNA icosahedron using fully atomistic molecular dynamics simulation in explicit water and ions. The major modes of internal motion have been identified using principal component analysis. We provide a quantitative estimate of the radius of gyration (Rg), solvent accessible surface area (SASA) and volume of the icosahedron which is essential to estimate its maximal cargo carrying capacity. Importantly, our simulation of gold nanoparticles (AuNP) encapsulated within DNA icosahedra revealed enhanced stability of the AuNP loaded DNA icosahedra compared to empty icosahedra. This is consistent with experimental results that show high yields of cargo-encapsulated DNA icosahedra that have led to its diverse applications for precision targeting. These studies reveal that the stabilizing interactions between the cargo and the DNA scaffold powerfully positions DNA polyhedra as targetable nanocapsules for payload delivery. These insights can be exploited for precise molecular display for diverse biological applications.

cond-mat.soft↗

arXiv2016

Nanoscale Structure and Elasticity of Pillared DNA Nanotubes

We present an atomistic model of pillared DNA nanotubes (DNTs) and their elastic properties which will facilitate further studies of these nanotubes in several important nanotechnological and biological applications. In particular, we introduce a computational design to create an atomistic model of a 6-helix DNT (6HB) along with its two variants, 6HB flanked symmetrically by two double helical DNA pillars (6HB+2) and 6HB flanked symmetrically by three double helical DNA pillars (6HB+3). Analysis of 200 ns all-atom simulation trajectories in the presence of explicit water and ions shows that these structures are stable and well behaved in all three geometries. Hydrogen bonding is well maintained for all variants of 6HB DNTs. We calculate the persistence length of these nanotubes from their equilibrium bend angle distributions. The values of persistence length are ~10 μm, which is 2 orders of magnitude larger than that of dsDNA. We also find a gradual increase of persistence length with an increasing number of pillars, in quantitative agreement with previous experimental findings. To have a quantitative understanding of the stretch modulus of these tubes we carried out nonequilibrium Steered Molecular Dynamics (SMD). The linear part of the force extension plot gives stretch modulus in the range of 6500 pN for 6HB without pillars which increases to 11,000 pN for tubes with three pillars. The values of the stretch modulus calculated from contour length distributions obtained from equilibrium MD simulations are similar to those obtained from nonequilibrium SMD simulations. The addition of pillars makes these DNTs very rigid.

cond-mat.soft↗

arXiv2014

Structure, stability and elasticity of DNA nanotube

DNA nanotubes are tubular structures composed of DNA crossover molecules. We present a bottom up approach for construction and characterization of these structures. Various possible topologies of nanotubes are constructed such as 6-helix, 8-helix and tri-tubes with different sequences and lengths. We have used fully atomistic molecular dynamics simulations to study the structure, stability and elasticity of these structures. Several nanosecond long MD simulations give the microscopic details about DNA nanotubes. Based on the structural analysis of simulation data, we show that 6-helix nanotubes are stable and maintain their tubular structure; while 8-helix nanotubes are flattened to stabilize themselves. We also comment on the sequence dependence and effect of overhangs. These structures are approximately four times more rigid having stretch modulus of ~4000 pN compared to the stretch modulus of 1000 pN of DNA double helix molecule of same length and sequence. The stretch moduli of these nanotubes are also three times larger than those of PX/JX crossover DNA molecules which have stretch modulus in the range of 1500-2000 pN. The calculated persistence length is in the range of few microns which is close to the reported experimental results on certain class of the DNA nanotubes.

cond-mat.soft↗