ASD ISM 2094 AI content filtering to block sensitive data exposure
What it says
Content filtering is implemented by AI applications to detect and block sensitive data exposure and improper output.
Quoted as published. Everything else on this page is written by Control Stack.
What it means
In plain EnglishAI apps use content filtering to prevent exposure of sensitive data and improper output.
AI tools need to filter the content they handle to make sure they don't accidentally spill sensitive information or produce unsuitable content. This is crucial because if confidential data gets out or inappropriate material is shown, it could lead to privacy breaches or harm the reputation of your business.
Free. Save your status, tasks and notes.
Why it matters
Without effective content filtering, sensitive data may be exposed, leading to privacy breaches and damaging reputational harm for your business.
How to implement
IT team should set up content filters in AI applications to block sensitive data exposure.
They can do this by configuring the AI software to recognise and mask or block data like financial info or patient records.
Managers should ensure training for staff about using AI tools carefully.
Organise sessions where staff learn the kind of data AI apps should not process, keeping them aware of potential risks.
Data protection officers should regularly review AI-generated outputs for improper content.
Use automated tools that flag risky outputs or manually check samples for compliance.
System owners should set rules for data types that AI apps can accept or refuse.
Work with the provider of your AI tools to clearly define sensitive data parameters that trigger filtering.
Procurement teams should check that any new AI tools include strong content filtering features before purchase.
Ask vendors during demos about how their filters work and what data they consider sensitive.
Keep it running
Regularly update AI content filtering settings to adapt to new data privacy laws and emerging threats, keeping your systems effective and compliant.
Evidence and audit
- Check 1 of 5What an auditor asks for
Ask for
the content filtering configuration guide: Request the document showing how AI applications are set to recognise and block sensitive info.
Look at
clear settings examples of blocked data types.
Good evidence
includes comprehensive filtering rules tailored to your organisation's needs.
- Check 2 of 5What an auditor asks for
Ask for
recent training records: Request details of any staff training sessions on the appropriate use of AI tools.
Look at
agendas covering data sensitivity and AI limitations. Good evidence includes participant lists and trainer qualifications.
- Check 3 of 5What an auditor asks for
Ask for
data output logs: Request logs of AI outputs that were flagged or blocked for review.
Look at
a variety of dates and scenarios demonstrating effective filtering.
Good evidence
includes notes on any corrections or updates made.
- Check 4 of 5What an auditor asks for
Ask for
the purchase documents of AI tools: Request contractual paperwork that details content filtering capabilities.
Look at
clauses that explain specific filtering mechanisms and obligations.
Good evidence
contains vendor commitments to protect sensitive data.
- Check 5 of 5What an auditor asks for
Ask for
audit findings on content filtering: Request the latest audit report reviewing AI applications' content filtering effectiveness.
Look at
unbiased third-party assessments focusing on filtering accuracy and improvement recommendations.
Good evidence
highlights strong performance and a plan to address any weaknesses.
Mappings
Closest match in ISO/IEC 27001: A.8.12, Data leakage prevention measures, which this control partially meets.
ISO/IEC 27001: 2 mappings
This control partially meets A.8.12.
ISM-2094 requires AI applications to implement content filtering to detect and block sensitive data exposure and improper output. ISO/IEC 27001:2022 Annex A 8.12 mandates data leakage prevention (DLP) measures across systems, networks and devices handling sensitive information, which can be implemented using AI-based content inspection and blocking. Annex A 8.12 is broader because it covers all DLP approaches (not just AI content filtering) and applies across wider technology surfaces than AI application outputs alone.
This control depends on A.5.13.
ISM-2094 requires AI applications to detect and block sensitive data exposure and improper output via content filtering. ISO/IEC 27001:2022 Annex A 5.13 requires information labelling aligned to the organisation's classification scheme. Effective AI content filtering typically depends on consistent labels/classification (e.g. sensitivity tags or categories) to tune detection rules and reduce false positives/negatives, so Annex A 5.13 provides a prerequisite foundation even though it does not mandate filtering.
Mappings show how controls relate across frameworks. They do not imply full equivalence or certification.