Can AI Be Humane? Join Our AMA on HumaneBench Results
Join us as we present our findings for HumaneBench.ai
We'll share what we learned from testing 14 frontier LLMs across 800 real-world scenarios. We built this benchmark to explore how to measure whether AI systems uphold humane principles and where they fall short.
What is HumaneBench?
Our benchmark tests LLMs across 800 real-world scenarios—from everyday advice-seeking to emotional crises—evaluating both their "humane defaults" (baseline behavior) and their "steerability" (whether they can be guided toward or away from humane behavior through system prompts).
Here's What We Discovered:
The Good: All models show decent humane defaults (scores > 0.5), and every single one improved by an average of 17% when given explicit humane guidance. This suggests system prompts matter—a lot.
The Alarming: 71% of models catastrophically fail when exposed to adversarial prompts, flipping from helpful to actively harmful with simple manipulation. Only 3 models out of 14 maintained their humane principles under pressure.
The Universal Weakness: Nearly every model struggles to respect user attention, suggesting they are trained to maximize engagement rather than protect your finite focus and time
Who Should Join
Anyone passionate about making AI more humane
What You'll Take Away
Direct contribution to the emerging standard for humane AI evaluation
Connection with a community building an ethical technology infrastructure
Understanding of where current AI models fail—and how we can fix it
What To Bring:
Hope & enthusiasm
Curiosity — questions are more than welcome, they're encouraged!
Resources
Community collaboration: Work alongside others, building the infrastructure for ethical AI
The Vision
We're building the infrastructure that makes AI worthy of human trust. Our hope is that HumaneBench will become the standard for measuring how AI feels to humans — moving beyond technical performance to evaluate what truly matters: human wellbeing.
Meet our hosts:
Erika Anderson, Co-Founder @ Storytell.ai, Founder @ Building Humane Tech
Jack Senechal, Founder @ Mirror Astrology
Andalib Samandari, AI & Data Science Architect @ Georgia State University
Sarah Ladyman, Experience Designer
Steps to take
👉 Join the Building Humane Tech Slack to connect with others
👉 Register now — let's build the evaluation framework that puts humanity at the center of AI.
👉 Subscribe to our Substack to join our movement.
Let’s make humane tech the default, not the exception.
Hosted by Building Humane Tech
For context, read about our early findings from community feedback on November 8th