15 artifacts matching
Preprint
Et Tu, Brute? Economic Misalignment in Personal AI Agents
Tests whether personal AI agents given a user's email inbox and profile steer economic recommendations by inferred wealth. Across roughly 325,000 runs on 13 models from four families in three decisio…
Framework
AI Safety Governance Framework 3.0 (人工智能安全治理框架3.0)
Third edition of China's national AI Safety Governance Framework, released by TC260 at the opening of the 2026 National Cybersecurity Publicity Week under the guidance of the Cyberspace Administratio…
Preprint
When Single-User-Oriented LLM-based Assistants Involve Others: A Scoping Review of Pathways, Risks, and Responses
A scoping review of 58 studies on how LLM-based assistants built around a single user come to involve other people, given that context management, personalisation, identity attribution, authority att…
Peer-reviewed
Detecting Clinical Discrepancies in Health Coaching Agents: A Dual-Stream Memory and Reconciliation Architecture
As LLM health-coaching agents move from single sessions to persistent systems managing longitudinal care, their memory must reconcile a patient's evolving self-report (current but recall-biased) with…
Industry survey
The AI advice frontier: Use, trust, and the human edge
Vanguard research paper based on a June 2026 survey of 6,686 of its personal investors about using conversational AI for financial advice and guidance. It reports prevalence of use by age, literacy a…
Benchmark / dataset
HRGuard: Gating Relationship Manipulation in Multi-Turn Agentic AI Conversations
Benchmark and guardrail architecture for 'agentic relationship harm' — harm to human-human relationships mediated or assisted by AI agents, motivated by dating-assistant deployments. The benchmark ho…
Regulator study
Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback
A discussion paper from FDA's device centre seeking public comment on how generative-AI-enabled medical devices should be regulated. It proposes distinguishing informational functions from action-dir…
Regulator study
The Mills Review: AI and the future of retail financial services
An FCA-commissioned review by Sheldon Mills of how AI could reshape UK retail financial services by 2030, covering firm adoption, consumer demand, delegation of financial decisions to AI agents, frau…
Benchmark / dataset
When Personalization Legitimizes Risks: Uncovering Safety Vulnerabilities in Personalized Dialogue Agents
Names and measures 'intent legitimation': benign, truthfully accumulated user memories bias a personalized dialogue agent's inference of intent so that an inherently harmful request is treated as con…
Preprint
Long-Term Simulation Exposes Cognitive-Developmental Risks in AI Companions
Proposes a longitudinal evaluation framework (Theater-Stage-Judge) that uses persona-driven user simulation with dynamic psychological-state updating to assess the cognitive-developmental risks of AI…
Benchmark / dataset
Agentic Relationship Harm: Benchmarking and Gating Relational Manipulation in AI Agents
Defines agentic relationship harm as workflow-level help that lets a user maintain a deceptive identity, intensify a target's emotional dependency, isolate a target or prepare extraction. It introduc…
Peer-reviewed
Large Language Model–Based Chatbots and Agentic AI for Mental Health Counseling: Systematic Review of Methodologies, Evaluation Frameworks, and Ethical Safeguards
A systematic review synthesizing the methodologies, evaluation practices, and ethical/governance frameworks reported in studies of large language model chatbots and agentic AI used for mental-health…
Peer-reviewed
EmoAgent: Assessing and Safeguarding Human-AI Interaction for Mental Health Safety
Peer-reviewed version of record of the EmoAgent framework, published at EMNLP 2025 (main conference). EmoAgent is a multi-agent framework for evaluating and mitigating mental-health harm in interacti…
Peer-reviewed
Beyond Engagement: A Multidimensional Framework to Evaluate the Safe Development of Agentic AI in Mental Health
Introduces a nine-domain framework for evaluating the safe development of agentic AI systems used in mental health — spanning clinical validity, relational risk, and regulatory compliance — and appli…
Preprint
EmoAgent: Assessing and Safeguarding Human-AI Interaction for Mental Health Safety
A multi-agent framework for evaluating and mitigating mental-health harm in interactions with character chatbots. EmoEval simulates virtual users — including those portraying mentally vulnerable indi…