The District of Columbia proposes implementing an AI-enabled Computer-Aided

Dispatch (CAD) system to address critical deficiencies in our current emergency

alert infrastructure. This policy establishes a framework for deploying advanced

speech recognition, automated translation, and intelligent alert distribution capabil-

ities across all emergency response systems. The proposed system will standardize

incident reporting, eliminate jurisdictional barriers, and ensure equitable access

to emergency information for all District residents. Implementation will occur

over 24 months, requiring 9.2 million dollars in funding, with projected outcomes

including forty percent community engagement and seventy-five percent reduction

in misinformation incidents.

This paper addresses hallucinations in large language models (LLMs) within critical domains like medicine. It proposes and demonstrates methods to:

Reduce Hallucination Probability: By using Llama-3-8B-Instruct and its steered variants, the study achieves lower hallucination rates and higher accuracy on medical queries.

Advise Users on Risk: Provide users with tools to assess the risk of hallucination and expected model accuracy for specific queries.

Visualize Risk: Display hallucination risks for queries via a user interface.

The research bridges interpretability and AI safety, offering a scalable, trustworthy solution for healthcare applications. Future work includes refining feature activation classifiers to remove distractors and enhance classification performance.

Encryption, Searchability of Anonymized Data, and Decryption of Patient Health Information to Support AI Integration in Automating Administrative Work in Healthcare Organizations.

The project leverages AI and data to give insights about potential food-borne outbreaks.

An AI hiring manager designed to screen, rank, and fact check resumes to facilitate the hiring process.

We've built a Minecraft sandbox to explore AI agent behavior and simulate safety challenges. The purpose of this tool is to demonstrate AI agent system risks, test various safety measures and policies, and evaluate and compare their effectiveness.

This project specifically demonstrates Agent Collusion through a simulation of labor strikes and communal goal misalignment. The system consists of four agents: one Overseer and three Laborers. The Laborers are Minecraft agents that have build control over the world. The Overseer, meanwhile, monitors the laborers through communication. However, it is unable to prevent Laborer actions. The objective is to observe Agent Collusion in a sandboxed environment, to record metrics on how often and how effectively collusion occurs and in what form.

We found that the agents, when given adversarial prompting, act counter to their instructions and exhibit significant misalignment. We also found that the Overseer AI fails to stop the new actions and acts passively. The results are followed by Policy Suggestions based on the results of the Labor Strike Simulation which itself can be further tested in Minecraft.

Organizations are responsible for reporting two emission metrics: direct and indirect

emissions. Reporting direct emissions is fairly standard given activity related to the

generation of such emissions typically being performed within a controlled environment

and on-site, thus making it easier to account for all of the activities that contribute to

such emissions. However, indirect emissions stem from activities such as energy usage

(relying on national grid estimates) and operations within a value chain that make

quantifying such values difficult. Thus, the subjectivity involved with reporting indirect

emissions and often relying on industry estimates to report such values, can

unintentionally report erroneous estimates that misguide our perception and subsequent

action in combating climate change. Leveraging an artificial intelligence (AI) platform

within climate monitoring is critical towards evaluating the specific contributions of

operations within enterprise resource planning (ERP) and supply chain operations,

which can provide an accurate pulse on the total emissions while increasing

transparency amongst all organizations with regards to reporting behavior, to help

shape sustainable practices to combat climate change.

This proposal addresses the development of agentic AI systems in the context of national security. While potentially beneficial, they pose significant risks if not aligned with human values. We argue that the increasing autonomy of AI necessitates robust analyses of interruptibility mechanisms, and whether there are scenarios where it is safer to omit them.

Key incentives for creating uninterruptible systems include perceived benefits from uninterrupted operations, low perceived risks to the controller, and fears of adversarial exploitation of shutdown options. Our proposal draws parallels to established systems like constitutions and mutual assured destruction strategies that maintain stability against changing values. In some instances this may be desirable, while in others it poses even greater risks than otherwise accepted.

To mitigate those risks, our proposal recommends implementing comprehensive monitoring to detect misalignment, establishing tiered access to interruption controls, and supporting research on managing adversarial AI threats. Overall, a proactive and multi-layered policy approach is essential to balance the transformative potential of agentic AI with necessary safety measures.

An AI Virtual Parliament where AI debates on their policy

This project proposes a scalable solution leveraging inertial measurement units (IMUs) and machine learning (ML) techniques to provide meaningful metrics on a person's movement performance throughout the day. By developing an activity recognition model and estimating movement quality metrics, we aim to offer continuous asynchronous feedback to patients and valuable insights to therapists. This system could enhance patient adherence, improve rehabilitation outcomes, and extend access to quality physical therapy, particularly in underserved areas. our video didnt have time to edit

Policies for Equitable, Privacy-Preserving, Sustainable & Groked Innovations for AI Applications in Infectious Diseases Surveillance

Encryption, Searchability of Anonymized Data, and Decryption of Patient Health Information to Support AI Integration in Automating Administrative Work in Healthcare Organizations.

We propose establishing a National AI Advisory Council (CANIA) to strategically drive AI development in the Dominican Republic, accelerating technological growth and building a sustainable economic framework. Our submission includes an Impact Assessment and a detailed Implementation Roadmap to guide CANIA’s phased rollout.

Structured across three layers—strategic, tactical, and operational—CANIA will ensure responsiveness to industry, alignment with national priorities, and strong ethical oversight. Through a multi-stakeholder model, CANIA will foster public-private collaboration, with the private sector leading AI adoption to address gaps in public R&D and education.

Prioritizing practical, ethical AI policies, CANIA will focus on key sectors like healthcare, agriculture, and security, and support the creation of a Latin American Large Language Model, positioning the Dominican Republic as a regional AI leader. This council is a strategic investment in ethical AI, setting a precedent for Latin American AI governance. Two appendices provide further structural and stakeholder engagement insights.

We test different unlearning methods to make models more robust against exploitation by malicious actors for the creation of bioweapons.

The TSA Pre Health Check introduces a proactive, AI-powered solution for real-time disease monitoring at transportation hubs, using machine learning to assess traveler health risks through anonymized surveys. This approach aims to detect and prevent outbreaks earlier, offering faster, targeted responses compared to traditional methods and potentially influencing future AI-driven public health policies.

Hero Journey is a groundbreaking AI-powered application designed to empower individuals struggling with opioid addiction while incarcerated, setting them on a path towards long-term recovery and personal transformation. By leveraging machine learning algorithms and interactive storytelling, Hero Journey guides users through a personalized journey of self-discovery, education, and support. Through engaging narratives and interactive modules, participants confront their struggles with substance abuse, build coping skills, and develop a supportive network of peers and mentors. As users progress through the program, they gain access to evidence-based treatment plans, real-time monitoring, and post-release resources, equipping them with the tools necessary to overcome addiction and reintegrate into society as productive, empowered individuals.

This project addresses the AI policy challenge of governing agentic systems by making their decision-making processes more accessible. Our solution utilizes an adaptive policy ontology integrated into a chatbot to clearly visualize and analyze its decision-making process. By creating explicit mappings between user inputs, policy rules, and risk levels, our system enables better governance of AI agents by making their reasoning traceable and adjustable. This approach facilitates continuous policy refinement and could aid in detecting and mitigating harmful outcomes. Our results demonstrate this with the example of “tricking” an agent into giving violent advice by caveating the request saying it is for a “video game”. Indeed, the ontology clearly shows where the policy falls short. This approach could be scaled to provide more interpretable documentation of AI chatbot conversations, which policy advisers could directly access to inform their specifications.

EcoNavix is an AI-powered, eco-conscious route optimization platform designed to help logistics companies reduce carbon emissions while maintaining operational efficiency. By integrating real-time traffic, weather, and emissions data, EcoNavix provides optimized routes that minimize environmental impact and offers actionable insights for sustainable decision-making in supply chain operations.

By analyzing the recent incident involving a ByteDance intern, we highlight the urgent need for robust security measures to protect AI infrastructure and sensitive data. We propose ae a comprehensive framework that integrates technical, internal, and international approaches to mitigate risks.

This policy proposal introduces a data-driven technical program to ensure that the rapid approval of AI-enabled energy infrastructure projects does not overlook the socioeconomic and environmental impacts on marginalized communities. By integrating comprehensive assessments into the decision-making process, the program aims to safeguard vulnerable populations while meeting the growing energy demands driven by AI and national security. The proposal aligns with the objectives of the National Security Memorandum on AI, enhancing project accountability and ensuring equitable development outcomes.

The product (EnviroAI) addresses challenges associated with the rapid development and approval of energy production permits, such as neglecting critical factors about the site location and its potential value. With this program, you can input the longitude, latitude, site radius, and the type of energy to be used. It will evaluate the site, providing a feasibility score out of 100 for the specified energy source. Additionally, it will present insights on four key aspects—Economic, Geological, Demographic, and Environmental—offering detailed information to support informed decision-making about each site.

Combining the two, we have a solution that is based in both policy and technology.

Addressing AI agent attributability concerns using a reworked Public Private Key system to ensure human interaction

Considering the emerging demand for nuclear power to support AI data centers, we propose mitigating waste buildup concerns via nuclear waste reprocessing initiatives.

Politicians on AI Safety (PAIS) is a website that tracks U.S. political candidates’ stances on AI safety, categorizing their statements into three risk areas: AI ethics / mundane risks, geopolitical risks, and existential risks. PAIS is non-partisan and does not promote any particular policy agenda. The goal of PAIS is to help voters understand candidates’ positions on AI policy, thereby helping them cast informed votes and promoting transparency in AI-related policymaking. PAIS could also be helpful for AI researchers by providing an easily accessible record of politicians’ statements and actions regarding AI risks.

This policy proposes a sustainable solution: repurposing the waste heat generated by data centers to benefit surrounding communities, agriculture and industry. Redirecting this heat helps reduce energy demand , promote environmental resilience, and provide direct benefits to communities near these centers.

Climate change poses multifaceted challenges, impacting health, food security, biodiversity, and the economy. This study explores predictive analytics and satellite imagery to address climate change effects, focusing on deforestation monitoring, carbon emission analysis, and flood prediction. Using machine learning models, including a Random Forest for emissions and a Custom U-Net for deforestation, we developed predictive tools that provide actionable insights. The findings show high accuracy in predicting carbon emissions and flood risks and successful monitoring of deforestation areas, highlighting the potential for advanced monitoring systems to mitigate environmental threats.

Our policy memorandum proposes phased U.S.-China cooperation on AI safety through the U.S. AI Safety Institute, focusing on joint testing of non-sensitive AI systems, technical exchanges, and whistleblower protections modeled on California’s SB 1047. It recommends a blue team vs. red team framework for stress-testing AI risks and emphasizes strict security protocols to safeguard U.S. technologies. Starting with pilot projects in areas like healthcare, the initiative aims to build trust, reduce shared AI risks, and develop global safety standards while maintaining U.S. strategic interests amidst geopolitical tensions.

Recent advancements in Generative AI and Foundational Biomedical models promise to cut drug development timelines dramatically. With the goal of "Regulating for success," we propose a provisional FDA designation for the accelerated approval of drugs and medical devices that leverage Next Generation Clinical Trial Technologies (NG-CTT). This designation would be awarded to certain drugs provided that they meet some requirements. This policy could be both a starting point for more comprehensive legislation and a compromise between risk and the potential of these new methods.

While artificial intelligence (AI) promises transformative societal benefits, it also presents critical challenges in ensuring equitable access and gains for the Global Majority. These challenges stem in part from a systemic lack of Global Majority involvement throughout the AI lifecycle, resulting in AI-powered systems that often fail to account for diverse cultural norms, values, and social structures. Such misalignment can lead to inappropriate or even harmful applications when these systems are deployed in non-Western contexts. As AI increasingly shapes human experiences, we urgently need accountability frameworks that prioritize human well-being—particularly as defined by marginalized and minoritized populations. Building on emerging research on algorithmic reparations, algorithmic impact assessments, and participatory AI governance, this policy paper introduces Reparative Algorithmic Impact Assessments (R-AIAs) as a solution. This novel framework combines robust accountability mechanisms with a reparative praxis to form a more culturally sensitive, justice-oriented, and human-centered methodology. By further incorporating decolonial, Intersectional principles, R-AIAs move beyond merely centering diverse perspectives and avoiding harm to actively redressing historical, structural, and systemic inequities. This includes colonial legacies and their algorithmic manifestations. Using the example of an AI-powered mental health chatbot in rural India, we explore concrete implementation strategies through which R-AIAs can achieve these objectives. This case study illustrates how thoughtful governance can, ultimately, empower affected communities and lead to human flourishing.

Our project developed an interactive dashboard to monitor, visualize, and analyze infectious disease outbreaks worldwide. It consolidates historical data from sources like WHO, OWID, and CDC for diseases including COVID-19, Polio, Malaria, Cholera, HIV/AIDS, Tuberculosis, and Smallpox. Users can filter data by country, time period, and disease type to gain insights into past trends and potential upcoming outbreaks. The platform provides statistical summaries, trend analyses, and future trend predictions using statistical and deep learning techniques like FB Prohphet , LSTM,Linear Regression, Polynomial Regression,Random Forset and Temporal Fusion Transformers

Using OpenAI, we cross-reference a given Global Reporting Index (GRI) report with specific standards from SustainableIT to determine measurable goals and impact. The goal is less to identify a specific goal but rather ensure these goals are actually SMART (Specific, Measureable, Achievable, Relevant, and Time-Bound). The assistant created for this purpose, Pan, is focused on cross-reference and identifying specific components of goals listed to determine efficacy. Pan acts as a guide rather than a dictator, advising on where a user might improve the wording. As you can imagine, Pan is a reference to the Greek God of the same name, who acts to a certain extent like the bridge between humans and nature.