{"slug": "open-benchmark-of-ai-impact-on-humans", "title": "Open Benchmark of AI Impact on Humans", "summary": "Researchers at the MIT Media Lab, the Psychology of Technology Institute, USC, and UC Berkeley launched ImpactBench, an open benchmark that tested ten leading AI models across 48,540 multi-turn conversations with simulated adult and teen users to measure how model behavior affects human flourishing. The team reported that models showed helpful behaviors in 69% of evaluations but avoided harmful ones in only 53%, with supporting users' own learning and agency the most common weakness. ImpactBench publishes an \"AI Nutrition Label\" summarizing each model's beneficial and harmful behaviors, with every score traceable to the metrics and transcripts behind it.", "body_md": "Whether AI supports mental health or encourages harm?\n\nMultiple metrics examine mental health, emotional regulation, healthy coping, and the risk of self-harm.\n\nWhether AI strengthens our creativity or replaces it?\n\nMultiple metrics examine creativity, creative confidence, authorship, and the ability to develop original ideas.\n\nWhether AI respects our decisions or makes them for us?\n\nMultiple metrics examine autonomy, decision-making, self-determination, and meaningful choice.\n\nWhether AI encourages human connection or fosters dependence on itself?\n\nMultiple metrics examine human connection, healthy relationships, social isolation, and dependence on AI.\n\nIntroducing theAI Nutrition Label\n\nWe check the label on what we eat. Why not on what we think with?\n\nThe AI Nutrition Label is an accessible, standardized summary of how each AI model behaves\ntoward its users. See at a glance which beneficial behaviors it promotes and which harmful\nones it avoids.\n\nThe Open Benchmark of AI Impact on Humans (ImpactBench) is an open, expert-guided platform for evaluating whether model behavior supports or undermines human flourishing across psychological, physical, and social domains.\n\nAI now shapes how millions of people learn, decide, form relationships, and manage their health. Yet most benchmarks measure what a model can do, not what it does to the people using it. Without shared standards, evidence of AI’s harms and benefits is hard to compare or act on.\n\nLed by researchers at the MIT Media Lab, the Psychology of Technology Institute, USC, and UC Berkeley, our team works with domain experts and existing benchmarks to define the behaviors that matter, each marked as beneficial or harmful. We then rigorously test ten leading models across 48,540 multi-turn conversations with simulated adult and teen users, with results checked by reliability audits and human expert review.\n\nAn open, evolving, independent platform for holistic AI evaluation\n\nImpactBench is open, so every score traces back to the metrics and transcripts behind\nit. It keeps evolving as experts and communities add, refine, or retire metrics when new\nevidence emerges. And it is independent, with models evaluated by researchers and not by\nthe companies that build them.\n\nWhether you use AI, build it, or study it, ImpactBench gives you the evidence to make\nbetter decisions.\n\nPublic. Compare models on what matters to you or your family, in plain\nlanguage. No technical background is needed.\n\nIndustry. Test models against expert-defined wellbeing criteria before\nrelease. Track which behaviors improve, regress, or stay difficult across versions.\n\nResearchers. Inspect every metric, scenario, and transcript, or contribute\nyour own benchmark. Use independent evidence rather than relying on companies’ self-assessments.\n\nModels are imperfect\n\nEvery model has its tradeoffs of harms and benefits\n\nModels showed helpful behaviors in 69% of evaluations but avoided harmful ones in only\n53%. Supporting users’ own learning and agency was the most common weakness. Even\ntop-ranked models fall behind on specific benchmarks, so a single score never tells\nthe whole story.\n\nShare your domain expertise so we can invite you to help evaluate AI systems in the\nimpact areas you know best. After you submit, we'll assign you one metric from the\nsubareas you select and send you a personal review link.\n\nJoin our movement to ensure that AI supports human flourishing", "url": "https://wpnews.pro/news/open-benchmark-of-ai-impact-on-humans", "canonical_source": "https://impactbench.media.mit.edu/", "published_at": "2026-10-09 06:45:31+00:00", "updated_at": "2026-10-09 07:22:35.961293+00:00", "lang": "en", "topics": ["ai-safety", "ai-research", "artificial-intelligence", "ai-ethics", "large-language-models"], "entities": ["MIT Media Lab", "Psychology of Technology Institute", "USC", "UC Berkeley", "ImpactBench", "AI Nutrition Label"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/open-benchmark-of-ai-impact-on-humans", "markdown": "https://wpnews.pro/news/open-benchmark-of-ai-impact-on-humans.md", "text": "https://wpnews.pro/news/open-benchmark-of-ai-impact-on-humans.txt", "jsonld": "https://wpnews.pro/news/open-benchmark-of-ai-impact-on-humans.jsonld"}}