Customer work
Turn a complaint into a useful internal summary without carrying the customer's name, account number or contact details into the prompt.
Prepare a clean working version before opening the AI tool.
Workplace AI training, built in Australia
AI training that changes what people do next
Saifety teaches staff how to use Microsoft Copilot, ChatGPT, Claude and coding assistants in workplace tasks. They practise the data decision, write a better brief and check the answer before it travels any further.
No credit card. Progress saved.No credit card. Choose the first module of any course and keep your progress.
Prompting Fundamentals
Live prompt practice
Challenge
Summarise a customer complaint for the team, without exposing anyone's personal details.
Your prompt
Summarise this complaint in 3 bullet points for support. Keep it factual. I've removed the name, email and account no.
No supported high risk identifiers found. Check the context before sending.
Rubric result
The brief is scored, not the model's answer.
5/6
Data safety
2/2
Task clarity
2/2
Useful limits
1/2
Live prompt exercise
5 of 6
The brief is checked for data safety, task clarity and useful limits.
- 7 courses
- Foundations and product training
- 93 lessons
- Written and interactive work
- Live practice
- A model response and fixed rubric
- Repeat anytime
- Completed lessons stay available
Built around ordinary work
The lesson starts where the job gets awkward
A useful course needs more than definitions. Each example gives the learner a recognisable job, one decision that matters and a sensible way to check the result.
Documents and policy
Pull actions from a long report while keeping the source, missing context and final approval visible to the person using it.
Ask for traceable notes, then check them against the original.
Spreadsheets
Question a dataset without assuming that a confident explanation proves the calculation, formula or source is right.
Check the numbers in the source file before sharing the finding.
Software
Give a coding assistant a bounded job with permitted files, expected tests and a review point before changes reach production.
Keep permissions narrow and inspect the diff yourself.
The habit before the prompt
Decide what the tool should see
The quality of a prompt begins with the material you choose to put in it. Remove what the task does not need, check the account and workspace, then brief the work.
A customer complaint needs a short internal summary
Source material stays offline
The original email includes a name, address, account number, exact transaction date and personal history.
The prepared task can be briefed
Customer A reported a duplicate charge last month. Summarise the issue for support in three factual points. Flag anything the next staff member still needs to confirm.
- Check the approved account first
- Replace names and unique references
- Keep only facts needed for the job
- Confirm the combination cannot identify someone
See what the wording produces in practice
Selected exercises send prepared material to a live model, show the answer and score the learner's brief against a fixed rubric. The learner can then revise the request and try again.
The response, safety scan and scoring model do separate jobs. A scan can still miss sensitive context, so preparation happens before submission.
Prepare the material
Remove names, identifiers and restricted context before the exercise begins.
Write the brief
State the job, audience, useful context, limits and the shape of the answer.
Inspect the result
Read what the model produced and check the prompt against a fixed scoring rubric.
Improve the request
Use the feedback to make the next prompt safer, clearer and easier to review.
One rollout, two useful views
Focused for staff and legible for the person running it
Learners see their next lesson, feedback and progress. Team owners see assignments, due dates and completion records without intruding on the learning experience.
The learner view
A clear next step and a reason to return
Progress is saved between sessions. Completed lessons can be repeated, and each course keeps its practical toolkit available after training.
Level 2
Capable
1,840 XP
640 XP to Confident
First steps
Safe hands
Prompt craft
Sharpshooter
Streak 7
7 day streak. Two more for a badge.
Team training
AI Safety Fundamentals
78%
complete
Priya N.
Complete
Tom R.
In progress
Aisha K.
In progress
Mark D.
Overdue
Updated just nowExport CSV
The team owner view
Enough detail to follow up properly
Assign courses, set dates, send reminders and export the training record. Billing follows active members and carries no seat minimum.
The course library
Start with judgement, then learn the product
The foundation courses travel across tools. The product courses cover the places where accounts, permissions, files and execution change the right way to work.
Make sound calls about workplace information, approved tools, unreliable answers, hostile instructions and mistakes that need to be reported.
- Workplace information
- Approved tools
Turn a loose request into a brief that another person can inspect. Choose the useful context, set the evidence boundary and define what a finished answer should contain.
- Task framing
- Context and evidence
Choose the Copilot surface that suits the job, work from permitted files and understand what permissions do before you ask for a summary or draft.
- Desktop Copilot
- Word and Outlook
Use Chat, Projects, Skills, Artifacts and Microsoft 365 with a clear job and source boundary. Practise Word edits, Excel calculations, Outlook drafts, PowerPoint updates and supervised Cowork tasks using a complete synthetic business case.
- Chat, Projects and Skills
- Search and Microsoft 365
Set repository instructions, permission boundaries and a testable task before an agent edits code. Review the diff and evidence before the work moves on.
- Repository context
- Planning and permissions
Use chats, Projects, Canvas, search and data analysis without treating every surface or account as the same workplace boundary.
- Chat and Projects
- Canvas
Give Codex a bounded software task, the repository rules and a clear verification path. Keep approval for risky actions and deployment with a person.
- Local workspace
- Cloud tasks
Completion evidence
A record that says what happened and nothing grander
Paid learners can receive a PDF completion certificate with a public verification reference. Team administrators can also see who completed assigned training, when they finished and the result Saifety recorded.
The credential records course completion. It does not provide professional accreditation, replace workplace policy or promise how someone will handle the next task.
Saifety.
Certificate of Completion
This is to certify that
John Smith
has successfully completed
AI Safety Fundamentals
Issued
20 Jun 2026
Score
96%
Credential
SAIF-A1B2-C3D4-5678-4000
Verify at saifety.au/verify
Straight answers
Questions from learners and team owners
The practical questions that tend to come up before a learner starts or a team commits.
What is an online AI training platform?
It gives people somewhere to learn workplace AI skills, make decisions and practise before the work matters. Saifety combines short lessons, interactive checks, live prompt exercises, practical reference tools and completion records in one place.
Is Saifety for individuals or teams?
Both. An individual can work through the courses at their own pace. A team can invite staff, assign courses, set due dates, monitor completion, send reminders and export its training record from one administration area.
Which AI tools does the training cover?
Saifety includes foundations that apply across tools, followed by dedicated courses for Microsoft Copilot, Claude and Claude Code, and ChatGPT and Codex. The product courses cover the decisions that change when the account, workspace, files or execution surface changes.
Does the training use a live AI model?
Yes. Selected sandbox exercises send a prompt that the learner has already stripped of identifying detail to an external AI provider. The model response, safety scan and scoring model remain separate so one output cannot quietly stand in for all three checks.
What does a Saifety certificate verify?
It confirms that a named learner completed a Saifety course, the issue date and the result recorded by the platform. It is a completion record. It does not provide professional accreditation or predict how the learner will handle every future task.
Can I try the platform before paying?
Yes. The first module of every course is available without a credit card. You can inspect the lesson and exercise format before choosing an individual or team plan.
How often is the AI training updated?
Every course is reviewed each quarter. Saifety publishes the review date, the substance of any change and whether earlier learners should refresh their completion. Small wording changes do not create unnecessary repeat work.
Let people try the work before you ask them to pay
The first module of every course is open without a credit card. Use it for your own training or inspect the experience before you roll Saifety out to a team.