IRAP
ISM June 2026 Changes: The New AI Controls Explained
21 June 2026 · Updated 16 Oct 2026
The ISM June 2026 changes add four controls aimed squarely at artificial intelligence, the first time the manual treats AI applications as their own attack surface. If a system holds OFFICIAL: Sensitive data or above and runs an AI feature, an IRAP assessment will now test against them. The control set has moved, and the evidence has to move with it.
What changed in the ISM in June 2026?
The headline change in June 2026 is AI. The Australian Signals Directorate added four new controls that name artificial intelligence directly, broadened a cryptography control to cover all data encrypted in transit, widened the software development guidance to include AI built code, and added a control on using cyber threat intelligence to detect incidents. None of it is cosmetic.
This release builds on a bigger structural shift from earlier in the year. In March 2026 ASD restructured the cyber security principles into six functions, govern, identify, protect, detect, respond and recover, aligning the ISM with the international NIST Cyber Security Framework. The Information Security Manual changes through the year, usually each quarter. June is one release of several, and the next one moved much further: the ISM September 2026 changes added 44 new controls, extending the AI set into agentic AI and adding a block on non-human identity.
What are the new ISM AI controls?
Four new controls answer that. They treat an AI application as something that can reach out to the open internet, take actions on its own, and drift from how it behaved last week, and each one closes a path an attacker or a misconfigured model could use.
| Control | What it asks for |
|---|---|
| ISM-2112 | AI applications that process classified data have their ability to directly access external public data sources disabled |
| ISM-2113 | AI applications are configured to flag organisationally defined risky actions for human approval before they run |
| ISM-2114 | Baselines of expected behaviour and performance for AI applications are established and monitored for unexpected deviations |
| ISM-2117 | Suitable AI models are used to augment the detection of cyber security events and the identification of cyber security incidents |
Read together, the intent is clear: keep a model that handles classified data off the open internet, put a human between the model and anything consequential, watch for the model behaving in a way it should not, and turn AI back on the defenders as a detection aid. The first three constrain the AI an organisation deploys, while the last one is permission to use AI in its own security operations.
Do the new ISM AI controls apply to our IRAP assessment?
It depends on the classification and whether the system actually uses AI. Every ISM control carries an applicability marking, non-classified, OFFICIAL: Sensitive, PROTECTED, SECRET or TOP SECRET, so the assessment selects and tailors the controls to the system in front of it. The control set is the same at OFFICIAL: Sensitive and PROTECTED; what changes higher up are the physical, personnel and network obligations.
For a system with no AI component, these controls are marked not applicable in the control matrix, with a justification, and the assessment moves on. Where teams come unstuck is the AI feature nobody scoped as AI: a support copilot, a document summariser, a model bolted onto a PROTECTED workload after the boundary was agreed. ISM-2112 names classified data systems directly, so the assessment will ask where the model can reach, and if the answer is the open internet, that is a finding.
What else changed in the ISM: cryptography, software development and threat intelligence?
Three changes are easy to miss behind the AI headline. An existing cryptography control, the use of an ASD Approved Cryptographic Protocol or a high assurance cryptographic protocol over network infrastructure, was amended to capture all scenarios where data is encrypted in transit, not only traffic crossing network infrastructure. That is a wider net than it reads.
The introduction to the software development fundamentals section was rewritten to state plainly that it applies to human, AI assisted, AI powered and AI driven software development. If a coding assistant writes part of a product, the secure development guidance still binds it. ASD also added a control, ISM-2116, recommending cyber threat intelligence services support the detection of events and the identification of incidents, the defensive sibling of the AI detection control ISM-2117.
Does an ISM update reset the IRAP clock?
No. A new ISM version does not retroactively invalidate a current authorisation. An IRAP assessment is point in time, measured against the version of the ISM used on the day. The report an organisation holds was assessed against the ISM as it stood then, and the authorising officer accepted that risk.
What the update changes is the next assessment. Under PSPF requirement 0109 a cloud service must have been IRAP assessed within the previous 24 months against the latest ISM at the time of assessment, so a reassessment from today is measured against the June 2026 manual, AI controls included. Material change forces a reassessment sooner, and this is the part teams underrate: adding an AI capability to an already assessed system is a material change. A copilot dropped into an authorised system is a change to the boundary, not a footnote. Where an AI feature has entered scope since the last report, Cybernion runs independent IRAP assessments against the current manual.
Frequently asked questions
Which ISM controls cover AI?
The June 2026 release added ISM-2112, ISM-2113, ISM-2114 and ISM-2117. They cover disabling direct external data access for AI that processes classified data, human approval for risky AI actions, behavioural baselining of AI applications, and using AI to augment threat detection.
Do the ISM AI controls apply if a system does not use AI?
No. If there is no AI component in scope, the ISM AI controls are marked not applicable in the control matrix, with a justification. They apply once an AI feature falls inside the assessment boundary.
Does a new ISM version invalidate an existing IRAP report?
No. The assessment is point in time against the ISM version used on the day, so a current authorisation stands. The next assessment is measured against the latest ISM, and PSPF requirement 0109 expects reassessment within 24 months.
Where can we read the ISM June 2026 changes?
ASD publishes a changes document alongside each ISM release on cyber.gov.au, listing every control added, amended or removed. The June 2026 changes document records the new AI controls and the cryptography amendment.
Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 Lead Auditor and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.
Sources:
- ISM June 2026 changes, ASD, June 2026
- Information Security Manual, ASD, June 2026
- The cyber security principles, ASD, March 2026
- Cloud assessment and authorisation, ASD, 2026
- Protective Security Policy Framework (Table 21, requirement 0109), 2024 release
Last updated: 4 September, 2026
