9 artifacts matching
Framework
Safe Participation Framework: Opportunity and Safety for the Next Generation in the Age of AI
Microsoft publishes a three-pillar corporate framework for young people's safety across its AI and online products, stating it intends the framework to inform emerging regulatory frameworks. The pill…
Preprint
What AI Benchmarks Actually Measure: Adapting Convergent and Discriminant Validity to Interrogate Fifty-Six AI Benchmarks
The paper adapts convergent and discriminant validity from the social sciences into a procedure for interrogating whether AI benchmarks measure the concepts they claim to measure, applying it to 56 c…
Preprint
When Seeing Overrides Knowing: Visual Dominance and Deferral-Based Method for Personalized Safety in VLMs
Extends personalized safety to vision-language models. MPS-Bench pairs 5,181 image-plus-query scenarios, built from 584 real-world images across 12 high-risk domains (including Health, Relationship,…
Lab publication
2026 Responsible AI Transparency Report
Microsoft's third annual responsible-AI transparency report, covering July 2025 to June 2026. One of its four 2026 trends is people turning to conversational AI for personal advice, health questions…
Preprint
Psychological Influences of Conversational AI: Research and Design Directions for Reducing Harm and Promoting Well-Being
Framework preprint examining how conversational AI systems affect users' psychological health, identifying benefits (information access, learning support) alongside risks including emotional entangle…
Peer-reviewed
Beyond the Single Turn: Reframing Refusals as Dynamic Experiences Embedded in the Context of Mental Health Support Interactions with LLMs
Argues that a refusal is not a single response to be scored for policy compliance but an experience that unfolds around the moment of refusal, and builds a five-phase framework from surveys and inter…
Preprint
Single-turn emergency psychiatric triage across 15 frontier AI chatbots
Benchmark study of how 15 frontier AI chatbots triage psychiatric urgency from a single user message. 112 clinical vignettes, each a one-message disclosure describing a change in behaviour, mental st…
Peer-reviewed
Public use of a generalist LLM chatbot for health queries
Peer-reviewed analysis of over 500,000 de-identified health-related Microsoft Copilot consumer conversations from January 2026 (N=617,827 after exclusions), classified with a 12-category hierarchical…
Industry survey
What Teens Say About AI Companionship: Key Findings from Youth Co-Design Workshops in India and Singapore
Findings from AI Companionship Youth Co-Design Sessions run with secondary-school students in India and Singapore, using live polling, ranking exercises and open prompts across four themes: AI in my…