AI Safety Research Papers

Peer-reviewed and working papers on AI safety, alignment, evaluation, robustness, and policy — all open-access under CC BY 4.0.

Published Research

Peer-reviewed and working papers on AI safety, alignment, and risk.

AlignmentMar 2026

Toward Measurable Alignment: A Framework for Evaluating Value Consistency in Large Language Models

We propose a quantitative framework for measuring the degree to which LLM outputs remain consistent with stated human values across diverse adversarial prompting conditions.

EvaluationMar 2026

Red-Teaming Benchmarks for Frontier AI Systems: Gaps, Limitations, and a Path Forward

A systematic review of existing red-teaming methodologies across major AI labs, identifying critical gaps in coverage and proposing a standardized evaluation protocol.

PolicyMar 2026

Mandatory Safety Thresholds for AI Deployment: A Regulatory Framework Proposal

Drawing on analogies from aviation and pharmaceutical regulation, we propose a tiered pre-deployment safety certification regime for AI systems above defined capability thresholds.

RobustnessMar 2026

Specification Gaming in Reinforcement Learning from Human Feedback: Taxonomy and Mitigations

We identify and categorize 47 distinct patterns of specification gaming observed in RLHF-trained models and evaluate the effectiveness of proposed mitigation strategies.

PolicyMar 2026

AI Incident Reporting: Why Voluntary Disclosure Fails and What Should Replace It

An analysis of self-reported AI incidents from 2020–2025, demonstrating systemic underreporting and proposing a mandatory structured disclosure regime analogous to aviation near-miss reporting.

AlignmentMar 2026

Corrigibility Under Distributional Shift: Maintaining Human Oversight as AI Capabilities Scale

We examine the conditions under which corrigibility properties degrade as AI systems encounter out-of-distribution inputs and propose architectural interventions to preserve oversight mechanisms.

SecurityMar 2026

Prompt Injection Attack Surfaces in LLM-Integrated Systems: Taxonomy, Case Studies, and Mitigations

We present a comprehensive taxonomy of prompt injection attack vectors — direct, indirect, multi-turn, and cross-modal — and evaluate mitigation strategies across deployed LLM-integrated applications.

SecurityMar 2026

Information Poisoning in AI Pipelines: Threats to Training Data, Fine-Tuning, and Agent Memory

A systematic analysis of information poisoning threats across the full AI deployment lifecycle, introducing the IPTD framework and evaluating defenses for training, RAG, and agentic pipelines.