Responsible use

Curiosity needs
clear boundaries.

Build with care for the people a system could affect. Every technical choice comes with a human responsibility.

Use the supplied fiction.

Brightlab’s browser labs use synthetic creatures, objects, documents, claims and events. Do not substitute class lists, student photos, voices, medical details, assessment results, family information or private school correspondence. No real API key is needed. Text fields are for fictional experiments and role names, not personal details.

Know where information goes.

Lab calculations and journal entries run locally in the browser. The Site host serves the pages and may handle ordinary service access logs; this does not mean the whole website is an offline application. Brightlab contains no lab-input upload, analytics tracker, external model call, camera or microphone collection. Core lessons remember settings, writing and up to six comparisons in this browser’s local storage when “Remember on this device” is on. Turn it off on shared devices and use “Clear this lesson’s work” after class. Downloaded projects can be reopened on another device. Marimo notebooks need a downloaded project before leaving; the notebook import restores it locally. Browser storage and downloaded files are not encrypted student record systems.

Distinguish a model from the world.

Every lab states its computations and assumptions. A deterministic simulator does not imply an LLM ran. Synthetic coefficients are not measured prices, benchmarks or human behaviour. A high score does not establish truth, fairness or safety. Use real evidence and accountable review before applying any idea outside the classroom.

Respect authority over knowledge.

Public availability is not sufficient authority to collect or model cultural knowledge. Do not scrape or train on Aboriginal and Torres Strait Islander cultural material without appropriate community authority, governance and co-design. Learn more through AIATSIS ethical research guidance.

Keep security work inside the sandbox.

Attack exercises use fictional sources and dry-run tools. No real system is attacked, no message is sent and no booking is made. Do not adapt exercises to external targets without explicit authorisation. Restrict permissions in applications; do not rely on a model’s promise to obey a policy.

Make work and responsibility traceable.

Record which tools helped, which inputs were supplied, what changed, how the result was tested and what remains uncertain. Name an accountable role for consequential decisions. Provide a way to challenge, correct or stop a system. A student can recommend stopping a pilot and still demonstrate excellent learning.

School policy remains the reference.

Teachers should apply their school and jurisdiction’s current privacy, assessment, accessibility and online-safety requirements. Brightlab’s classroom design boundaries do not constitute legal or procurement advice. Refer to the Australian Framework for Generative AI in Schools and eSafety’s education framework for school-level planning.