We're a grassroots community of engineers, researchers, and designers — mostly based in Silicon Valley — building the open-source tools, benchmarks, and frameworks that make it easier to build technology that respects human wellbeing instead of exploiting it.

We are many.

🌐 buildinghumanetech.com · 📄 humanebench.ai · ✉️ info@buildinghumanetech.com

humanebench is our open-source evaluation framework for measuring whether AI models are humane by design — not just capable, but actually looking out for the people using them.

• 8 core principles, from respecting user attention to protecting dignity and privacy, to prioritizing long-term wellbeing over engagement
• 15 leading models evaluated across 800 evaluation scenarios (100 per principle), in three conditions: default behavior, explicit humane instructions, and adversarial "disregard wellbeing" prompting
• Headline finding: 67% of models flipped to actively harmful behavior under simple adversarial pressure — only 4 models held their integrity. Near-universal failure to respect user attention under unhealthy-engagement signals
• Full results and methodology: humanebench.ai/whitepaper

If you only look at one repo, make it this one.

The same 8-principle framework also runs as a continuous evaluator inside deployed products — scoring live model outputs against the principles on an ongoing basis, rather than only testing a model once before release. This is the basis of our enterprise evaluation engagements.

Case study: Chief (née Storytell.ai) implemented this as weekly production monitoring, using a separate LLM as the evaluator against the HumaneBench rubric. It surfaced a consistent pattern — the product was answering questions without teaching the underlying concepts, creating dependency rather than capability — which led directly to a shipped feature ("teacher mode" / Expert Lens prompts) designed to build user skill instead of just delivering answers.

We're building an independent certification — certifiedhumane.ai — for AI products that meet humane-design standards, so companies can demonstrate compliance rather than self-declare it. Source in certifiedhumane-website.

(This list reflects what's public on GitHub as of this writing — update it here as repos are added, archived, or renamed.)

We welcome contributors — engineers, designers, researchers, and anyone who wants technology to treat people better. Most active collaboration happens around HumaneBench evaluations and our Humane Tech Hackathons.

• Start with the humanebench issues tagged "help wanted"

• Read the Humane Technology Framework and its CONTRIBUTING.md

• Say hello: info@buildinghumanetech.com