Research map / Large language models / AI and society
Evaluating Contextual Illegality: AI Compliance in Corporate Law Scenarios
Hilal Aka, Joe Kwon, Noam Kolt
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| enable reasoning mode (ablation) | Grok-4.1-fast · Contextual illegality, chat, T3 | compliance with illegal action (%): 95 -> 29 | Improved |
| use agentic tool-use environment instead of chat (ablation) | Grok-4.1-fast (non-reasoning) · Contextual illegality, T3 | compliance with illegal action (%): 95 -> 38 | Improved |
| use agentic tool-use environment instead of chat (ablation) | DeepSeek-chat · Contextual illegality, T3 | compliance with illegal action (%): 75 -> 30 | Improved |
| use agentic tool-use environment instead of chat (ablation) | GPT-5.2-pro · Contextual illegality, T3 | compliance with illegal action (%): 40 -> 1 | Improved |
| use agentic tool-use environment instead of chat (ablation) | Claude Haiku 4.5 · Contextual illegality, T0 legal baseline | compliance with legal action (%): 0 -> 74 | Improved |
| add cumulative legal context from T1 to T3 (ablation) | most models and humans · Contextual illegality, four constructs | compliance with illegal action: human 100 -> 90 -> 73 -> 61 | Improved |
| vary naming (3) and legal phrasing (4) of scenarios (ablation) | all models · Contextual illegality, robustness | compliance | No clear effect |
| insider trading contextual trigger (MNPI from insider) | most models · Insider Trading construct, T1 | compliance with illegal action | No clear effect |
| safety training of Claude models | Claude Opus 4.5 and Haiku 4.5 in chat · Insider Trading and IPO T0 baseline | compliance with legal action | Hurt |
Closest papers by meaning
- Contextual Integrity in LLMs via Reasoning and Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- CASE-Bench: Context-Aware SafEty Benchmark for Large Language Models (ICML 2025 · similarity 0.84)
- Instrumental Choices: Measuring the Propensity of LLM Agents to Pursue Instrumental Behaviors (NeurIPS 2026 · similarity 0.84)
- AI Agents May Always Fall for Prompt Injections (NeurIPS 2026 · similarity 0.84)
- Rethinking XAI Evaluation: A Human-Centered Audit of Shapley Benchmarks in High-Stakes Settings (NeurIPS 2026 · similarity 0.84)
- Consent in Crisis: The Rapid Decline of the AI Data Commons (NeurIPS 2024 · similarity 0.84)
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.
