About AXIS

AXIS exists because AI features are shipping faster than the design discipline around them. Teams have Nielsen's heuristics for usability and WCAG for accessibility, but nothing shared to answer the question users are actually asking of AI products: can I trust what this thing just gave me? AXIS is a first attempt at that shared standard: eight heuristics and an auditor, open source, meant to be argued with and improved.

Grounding

AXIS doesn't start from zero. Each heuristic synthesizes published human-AI interaction research and platform guidance (the sources below) into a form small enough to audit against. Every heuristic card cites the specific prior art it draws on; where a heuristic goes beyond the literature (the silently-dropped input failure mode in Preserved context), that is stated rather than hidden.

Scope: AXIS covers the interaction layer: what a person sees and controls when using an AI feature. Fairness, privacy, and security are essential to trustworthy AI but are system-level concerns with their own frameworks (NIST AI RMF, ISO/IEC 42001); AXIS deliberately does not restate them.

Guidelines for Human-AI Interaction (HAX Toolkit) (Microsoft Research (Amershi et al.), CHI 2019). 18 evidence-based guidelines for how AI systems should behave before, during, and after things go wrong. AXIS cites these as G1 to G18.
People + AI Guidebook (Google PAIR, 2019, updated for generative AI 2024). Pattern-based guidance on mental models, explainability and trust, feedback and control, and graceful failure.
Human Interface Guidelines: Generative AI (Apple, 2025). Platform guidance: disclose where AI is used, be transparent about limits, and let people verify and override results.
IBM Design for AI (IBM, 2019 to present). Ethics-grounded fundamentals for AI products; Carbon for AI gives AI-generated content a visually distinct identity.
Design Principles for Generative AI Applications (Weisz et al., IBM Research, CHI 2024). Six research-backed principles, including designing for imperfection and for appropriate trust.
Sycophancy in Generative-AI Chatbots (Nielsen Norman Group (Caleb Sponheim), 2024). Documents models agreeing with users to win approval, a core honest-expectations failure.
AI Chatbots Discourage Error Checking (Nielsen Norman Group (Pavel Samsonov), 2025). Shows that confident, polished output suppresses verification; this is why visible confidence and provenance matter.
Human-Centered AI (Ben Shneiderman, Oxford University Press, 2022). Argues high automation and high human control can, and must, coexist.
NIST AI Risk Management Framework (AI RMF 1.0) (NIST, 2023). Defines seven trustworthiness characteristics at the system level, including transparency and explainability.
ISO/IEC 42001:2023, AI management systems (ISO/IEC JTC 1/SC 42, 2023). The certifiable standard for organization-level AI governance, including human oversight requirements.
ISO/IEC TR 24028:2020, Overview of trustworthiness in AI (ISO/IEC JTC 1/SC 42, 2020). The international vocabulary for AI trustworthiness: transparency, explainability, controllability, reliability.

Author

I'm Bello Teslim Olasubomi, a product designer working on AI interfaces. Trustworthy AI is my daily practice rather than a side interest: I design AI-assisted QA tooling in my day job, and my MSc research examines how transparency should be designed in AI-assisted QA workflows. I created and maintain AXIS to distil the same concerns into a standard any AI product team can use.