Research map / Large language models / AI and society
AI Risk Profiles: A Standards Proposal for Pre-deployment AI Risk Disclosures
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a risk-centric "Risk Profiles" standard with a nine-category AI risk taxonomy and a template-based methodology for pre-deployment risk disclosure, illustrated with profiles of five popular AI systems.
Problem
Current AI disclosure conventions such as model cards and technical reports do not standardize risk disclosure or give decision-makers across the AI value chain a common risk framing.
Approach
A high-level risk taxonomy (abuse & misuse, compliance, environmental & societal impact, explainability & transparency, fairness & bias, long-term & existential risk, performance & robustness, privacy, security) plus a standardized report template that pairs identified risk scenarios with developer mitigations, evaluations, and certifications, built from public information.
Key results
- Five Risk Profiles (Claude, GPT APIs, Microsoft Copilot, GitHub Copilot, Midjourney) were produced from publicly available information using the proposed template.
- The taxonomy has nine high-level risk categories, and the risk and mitigation summary table indicates which apply and whether the developer provides mitigations.
- The authors check compliance against 8 standards and regulations, including GDPR, CCPA, and SOC II.
Limitations
- Binarized risk presence treats all risks as equally probable and impactful, which may mislead consumers.
- Risk scenario discovery relies on subjective subject-matter expertise with latitude for bias.
- Evaluations are limited by the nascent state of AI benchmarks and reliance on developer-published results.
Experimental setup
Datasets: publicly available system documentation developer evaluations academic benchmarks (e.g., MMLU)
Benchmarks: MMLU LMSys chatbot arena rankings
Metrics: risk presence per taxonomy category mitigation presence per risk
Baselines: Model Cards developer technical reports
Builds on: Model Cards EU AI Act (draft) algorithmic impact assessments
Closest papers by meaning
- Market Incentives for AI Safety Investment (NeurIPS 2026 · similarity 0.82)
- Position: Require Frontier AI Labs To Release Small "Analog" Models (NeurIPS 2025 · similarity 0.81)
- Position: Standardization of Behavioral Use Clauses is Necessary for the Adoption of Responsible Licensing of AI (ICML 2024 · similarity 0.81)
- Defining Operational Conditions for Safety-Critical AI-Based Systems from Data (NeurIPS 2026 · similarity 0.81)
- EVALUATION CARDS: An Interpretive Layer for AI Evaluation Reporting (NeurIPS 2026 · similarity 0.81)
- SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior (ICML 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.