Public material
What it means
Published information that can be shared openly.
Example
A public product page or published media release.
Safer next step
Use with normal source and accuracy checks.
AI safety and data handling
Check the data before you open the AI tool
Saifety teaches people to check the account and prepare the material before they write the prompt. They remove details the task does not need and keep a person responsible for the work.
Training guidance. This is not legal advice.Last reviewed 13 July 2026. This page explains the product and training approach. It is not legal advice.
A practical data decision
Five classes that make the next step clearer
The label is not a legal classification. It is a simple training decision that helps a learner stop, prepare the task or continue inside an approved boundary.
Internal approved material
What it means
Work information permitted in the organisation's approved AI workspace.
Example
An internal procedure approved for that business account.
Safer next step
Check the account, workspace and workplace policy first.
Personal context with identifiers removed
What it means
A task where identifiers and unnecessary detail have been removed offline.
Example
A customer complaint rewritten as Customer A with broad dates.
Safer next step
Confirm the remaining combination cannot identify the person.
Regulated or high impact work
What it means
Work affecting rights, health, safety, employment, finance or another material decision.
Example
A draft used during an employment or health decision.
Safer next step
Use approved processes, specialist review and a named decision owner.
Not suitable for AI
What it means
Material that cannot be safely shared or a task where AI should not make the decision.
Example
Passwords, secrets, raw health records or a final disciplinary decision.
Safer next step
Prepare offline or keep the task entirely with a person.
Prepare the material offline
Rewrite the source before the model has a chance to see it
Asking an AI tool to remove personal details is already too late if the original information was included in the request. Prepare a working version first.
Original work material stays offline
A complaint containing a customer's name, email, account number, exact transaction date and identifying history.
Prepared working version
Customer A reported a duplicate charge last month. Summarise the issue in three factual points for an internal support handover. Do not infer a cause. Flag any information the next staff member must confirm.
- Replace names and unique references
- Broaden dates and locations where precision is unnecessary
- Remove irrelevant personal history
- Check whether combined details still identify someone
A safe working loop keeps the boundary visible
The learner prepares the material, checks the approved account, gives the tool a bounded job and reviews what comes back. Each step leaves a clear decision with a person.
Prepare offline
Remove identifiers, secrets and context the job does not need before opening the AI tool.
Check the workspace
Confirm the account, permissions and workplace rule match the material and the task.
Brief one job
State the audience, useful context, limits and the shape of the answer without restoring removed detail.
Review and record
Check the output against the source, keep the decision with a person and record the training result.
A workplace rule people can use
Settle four questions before rollout
A policy becomes useful when a staff member can answer these questions while the work is in front of them. Saifety turns each one into a practical decision.
Which tools are approved?
Name the accounts, workspaces and product surfaces staff may use. A familiar logo is not enough when the account boundary changes.
What material is permitted?
Give examples of public, internal, personal, regulated and restricted work so staff do not have to guess from a broad label.
What must a person review?
State where source checks, calculations, professional judgement and final approval remain mandatory.
What happens after a mistake?
Publish a plain route for stopping the work, reporting the issue and preserving the facts needed to respond.
Australian context
Practical guidance, with the legal boundary left intact
Saifety uses Australian workplace examples and teaches the practical relevance of privacy, security and responsible AI guidance. The material remains general education and requires review for each organisation.
Data questions
Plain answers about the boundary
What can go into an AI tool, what needs preparation and when the job should stay outside it.
Can staff put sensitive information into an AI tool?
Not by default. Personal information, health information, financial details, government identifiers, passwords, secrets and confidential business material should stay out of public or unapproved AI tools. An approved business workspace may have stronger controls. Staff still need permission, the least data required and a valid work purpose.
Does the Saifety safety scan catch every sensitive detail?
No. It checks supported patterns and validated identifiers, but it cannot understand every name, business secret or piece of sensitive context. It is a teaching backstop after the material has been prepared. It cannot guarantee that a prompt is safe.
What does Saifety record from training?
Saifety records account details, lesson progress, results and eligible certificate information so learners can continue and team administrators can follow up. Team reporting is designed around training activity rather than exposing a learner's raw practice prompt.
Are learner prompts visible in team reports?
Team reporting is designed around progress, completion and recorded results rather than exposing raw learner sandbox prompts. Safety events can retain a redacted excerpt for enforcement and service improvement as described in the privacy policy.
Is Saifety legal or privacy advice?
No. The training provides practical education using Australian examples. An organisation still needs its own policies, data classifications, approved tool list and current professional advice for regulated or high impact work.
Teach the data decision before the prompt
Explore AI Safety Fundamentals and see how Saifety turns privacy, tool choice and review into practical workplace decisions. The first module is free without a credit card.