Plan for twice-yearly cycles
Vulnerability assessments and penetration tests, DCSync permission reviews and KRBTGT password changes move from yearly to at least every six months. Book the extra round now.

ASD's eight prioritised mitigation strategies, at three maturity levels.
152 requirements

The Australian Government's Information Security Manual, guideline by guideline.
1,143 controls

Everything ASD added, changed and rescinded in the September 2026 Information Security Manual, in plain English. Open any control for its full guidance.
Reviewed 15 November 2026 · Compared with the June 2026 ISM · 1,143 controls in force
ASD added 44 controls, changed 23 and rescinded 2. Credentials and sign-in, network devices and AI agents get the most new controls, and three yearly tasks now come round every six months.
The release brings the ISM to 1,143 controls. Thirteen new controls tighten credentials, sign-in, sessions and third-party app consent. Seven cover AI agents: each needs its own identity, an entry in an AI agent register, only the tools and permissions its purpose needs, and central logging of what it does. Networking gains eight, five of them MACsec settings. Another 82 controls are reworded. Many now say 'human users', because ASD's glossary now defines a user as a human or non-human actor, such as a service, application, workload or AI agent.
What this release means if you assess, run or build systems against the ISM.
Vulnerability assessments and penetration tests, DCSync permission reviews and KRBTGT password changes move from yearly to at least every six months. Book the extra round now.
Device-bound tokens, session revocation, administrator-only OAuth consent and switching off device code sign-in where it is not needed all target ways attackers gain access without stealing a password.
Record each AI agent in a register with its owner, identities, credentials and access, and cut its tools and permissions to what its purpose needs. AI applications must also get human approval before sensitive or high-impact actions.
802.1X controls now cover wired networks, four Active Directory controls cover every AD CS server rather than only CA servers, and two credential controls widen from networks to systems. Identity and access controls that now say 'users', such as ISM-0414 and ISM-0430, also cover services, workloads and AI agents.
Grouped by theme. Open a control for its plain-English guidance, steps and audit checks.
Every AI agent gets its own identity and an entry in an AI agent register that is regularly verified. Agentic AI applications get only the tools and permissions their purpose needs, and each tool call is limited to what both the user and the agent's task allow. Retrieved content stays untrusted and can never override instructions, permissions or approval steps. Every tool call, external request and output is logged centrally.
Applications and workloads should use short-lived credentials where they can, and no credential is shared with another application or across development, testing, staging and production. Static credentials live in a secrets manager, are changed when exposed and are revoked when no longer needed. Unneeded user account credentials are revoked too.
Only an authorised administrator can consent to a third-party OAuth application. Consents are reviewed at least every six months, and unused apps and excess permissions are removed. Consent and token activity is logged centrally. Device code sign-in stays off unless it is needed, and then only authorised accounts on managed devices can use it.
Tokens and session cookies are cryptographically bound to the device they were issued on. Sessions and tokens are revoked after a credential reset or re-enrolment, a known or suspected compromise, a device falling out of compliance or a high-risk sign-in. Access decisions are risk-based, using contextual signals.
Service provider access is limited to approved tools, source addresses and time windows. The organisation logs that access itself, where the provider cannot change or delete the logs, and checks the logs promptly for anything unexpected. Only authorised remote monitoring and management (RMM) and remote access tools are allowed, and gateways block the rest.
Web enrolment is turned off unless it is needed, and where it is on, it requires HTTPS and Extended Protection for Authentication. Certificate templates are reviewed at least every three months to find and fix settings that could allow privilege escalation or unauthorised enrolment. Every enrolment request, whether it succeeds or fails, and every change to templates or AD CS settings is logged centrally.
Management interfaces on IT equipment are reachable only from a dedicated, segregated management network. Network device firmware and running configurations are checked against an approved known-good baseline at least monthly, after patching and when a device behaves oddly. Unneeded components, services and functions are disabled or removed, and five controls set how MACsec is configured.
Driver signatures are checked before drivers load. Only privileged users who need it for their duties can install, load or change kernel-mode code such as drivers, kernel modules and extensions. WMI activity, including new permanent event subscriptions, is logged centrally.
Backups are immutable for their whole retention period, backup infrastructure is kept apart from production with its own administrator sign-in, and threat hunting, guided by current strategic and sector-specific threat intelligence, happens at least every three months.
Dependencies are pinned to approved versions in source code, and software is built reproducibly, so an independent party can confirm a release came from its stated source.
Before changing user account or banking details or making a payment, staff confirm who is asking through a pre-agreed authentication method or a separate trusted channel, because voice and video can be convincingly faked.
Controls whose requirement changed. Open "Compare ASD's wording" to see the statement with removed words struck through and added words highlighted.
Three recurring tasks now come round twice as often.
Vulnerability assessments and penetration tests now repeat at least every six months after deployment, not yearly.
Vulnerability assessments and penetration tests are conducted for systems prior to their deployment, including prior to the deployment of significant changes, and at least Removed: annually Added: every six months thereafter.
The KRBTGT password is changed twice when it has not changed in the past six months. It was 12 months.
Credentials for the Kerberos Key Distribution Center's service account (KRBTGT) are changed twice, allowing for replication to all Microsoft AD DS domain controllers in-between each change, if the domain has been directly compromised, the domain is suspected of being compromised or they have not been changed in the past Removed: 12 Added: six months.
DCSync permissions are reviewed at least every six months, not yearly.
User accounts with DCSync permissions are reviewed at least Removed: annually, Added: every six months, and those without an ongoing requirement for the permissions have them removed.
Monitoring after an intrusion is cleaned up now runs until there is evidence the intruder is gone and has not regained access, instead of at least seven days of full traffic capture. Remediation is planned on trusted systems and sequenced to limit re-compromise, rather than done in one planned outage.
At least seven days of full traffic capture after remediation is replaced by enhanced monitoring that continues until there is evidence the intruder is gone and has not regained access.
Following intrusion remediation activities, Removed: full network traffic Added: enhanced monitoring is Removed: captured for at least seven days and analysed to determine whether Added: conducted until there is sufficient evidence-based confidence that malicious actors have been Removed: successfully removed Added: eradicated from Removed: the system.Added: a system and have not re-established access.
Remediation is planned and coordinated on trusted systems separate from the compromised one, not just on any separate system.
Planning and coordination of intrusion remediation activities are conducted Removed: on a Added: using trusted systems separate Removed: system to that which has been compromised.Added: from a compromised system.
The single planned outage is gone. Remediation is coordinated and sequenced to limit re-compromise, balanced against operational risk and business continuity.
Removed: To the extent possible, all intrusion Added: Intrusion remediation activities are Removed: conducted in a coordinated Removed: manner during the same planned outage.Added: and sequenced to minimise opportunities for re-compromise of a system while balancing operational risk and business continuity requirements.
The 802.1X controls now cover wired networks as well as wireless. ISM-1321 adds key exchange, and the certificate controls now say X.509 certificates.
Covers key exchange as well as mutual authentication, and no longer names X.509 certificates, which ISM-1323 now requires.
802.1X Removed: authentication with Removed: EAP-TLS, using X.509 certificates, Added: EAP-TLS is used for mutual Removed: authentication; Added: authentication and key exchange; with all other EAP methods disabled on supplicants and authentication servers.
Applies to every 802.1X implementation, wired or wireless, not only wireless networks. Wireless access points are no longer listed separately.
Evaluated supplicants, Removed: authenticators, wireless access points Added: authenticators and authentication servers are used Removed: in wireless networks.Added: for 802.1X authentication implementations.
Applies to any network using 802.1X, wired or wireless, and names X.509 certificates.
Removed: Certificates Added: X.509 certificates are required for devices and Added: human users Removed: accessing wireless networks.Added: authenticating to networks using 802.1X.
Names X.509 certificates.
Removed: Certificates Added: X.509 certificates are generated using an evaluated certificate authority or hardware security module.
Names X.509 certificates, and only certificates issued to human users need authentication before use.
Removed: Certificates Added: X.509 certificates are protected by logical and physical access Removed: controls, encryption, Added: controls and Removed: user authentication.Added: encryption, with those issued to human users requiring authentication before use.
Four Active Directory controls that named AD CS CA servers now cover every AD CS server.
Covers all Microsoft AD CS servers, not only CA servers.
Security-relevant events for Microsoft AD DS domain controllers, Microsoft AD CS Removed: CA servers, Microsoft AD FS servers and Microsoft Entra Connect servers are centrally logged.
Covers all Microsoft AD CS servers, not only CA servers.
Microsoft AD DS domain controllers, Microsoft AD CS Removed: CA servers, Microsoft AD FS servers and Microsoft Entra Connect servers are only used for their designed role and no other applications or services are installed, unless they are security related.
Covers all Microsoft AD CS servers, not only CA servers.
Access to Microsoft AD DS domain controllers, Microsoft AD CS Removed: CA servers, Microsoft AD FS servers and Microsoft Entra Connect servers is limited to privileged users that require access.
Covers all Microsoft AD CS servers, not only CA servers.
Backups of Microsoft AD DS domain controllers, Microsoft AD CS Removed: CA servers, Microsoft AD FS servers and Microsoft Entra Connect servers are encrypted using ASD-approved cryptography, stored securely and only accessible to backup administrator accounts.
Wider scope for finding, protecting, issuing and changing credentials.
Credentials found stored in the clear anywhere on a system now have to be changed, not only those found on networks.
Credentials for user accounts are changed if:
• they are compromised Removed: • they are Added: or suspected of being compromised
• they are discovered stored on Removed: networks Added: systems in the clear
• they are discovered being transferred across networks in the clear
• membership of a shared user account changes.
Identity checks now cover first requests, resets, temporary disabling, and enrolment or re-enrolment of any credential.
Removed: Users Added: Human users provide sufficient evidence to verify their identity when Added: first requesting Removed: new Added: credentials, when requesting the reset of any credentials, when requesting the temporary disabling of any credentials, and when requesting the enrolment or re-enrolment of any credentials.
The monthly scan for credentials stored in the clear covers systems, not only networks.
Removed: Networks Added: Systems are scanned at least monthly to identify any credentials that are being stored in the clear.
Commit scanning must catch credentials as well as keys and secrets.
Scanning is used during commits to identify plain text or encoded Removed: secrets Added: credentials, keys and Removed: keys, Added: secrets, which are then blocked from being stored in the authoritative source for software.
TLS key establishment, service provider incidents, the developer skills register and human approval for AI applications.
Requires the ephemeral variant of DH or ECDH, absorbing ISM-1448, which is rescinded.
Added: The ephemeral variant of DH or ECDH is used for key establishment of TLS connections.
Unauthorised access to, or administration of, your systems by a service provider is now treated as a cyber security incident, as well as being reported to you straight away.
If an organisation's systems are accessed or administered by a service provider in an unauthorised manner, Added: it is treated as a cyber security incident and the organisation is immediately notified.
Now includes developing the register in the first place, not only implementing and maintaining it.
A software developer cyber security knowledge and skills register is Added: developed, implemented and maintained.
Human approval is now required before any sensitive or high-impact action. Before, approval applied to risky actions the organisation chose to define.
AI applications are configured to Removed: flag organisationally defined risky actions for Added: require human approval Removed: prior to their execution.Added: before executing sensitive or high-impact actions.
Changes ASD lists as wording only, keeping each control's intent. Check any policy or plan that quotes these controls, and compare the wording wherever it is shown.
When SSH-agent or similar key caching applications are used, Removed: it is limited to workstations and servers with Added: cached private keys have a maximum lifetime of four hours and, where applicable, screen locks Removed: and key caches that are Removed: set to expire within four hours of inactivity.Added: used on workstations and servers.
Trustworthy sources for SECRET and TOP SECRET systems are limited to people and Removed: services Added: systems that have been verified and authorised as such by the chief information security officer.
Removed: User Added: Interactive user sessions are terminated and workstations are restarted at least daily.
System Removed: administrators perform system administration activities Added: are performed in accordance with the system's change and configuration management plan.
Database servers are placed on a different network segment to Removed: user workstations.
Removed: Where supported, 64-bit versions of operating systems are used.
All queries to databases from Removed: software that are initiated by users, Added: software, and any resulting crash or error messages, are centrally logged.
Removed: Personnel Added: Users granted access to systems and their resources are uniquely identifiable.
The use of shared user accounts is strictly controlled, and Removed: personnel using Added: users of such accounts are uniquely identifiable.
Access to systems and their resources are removed or suspended the same day Removed: personnel Added: users no longer have a legitimate requirement for access.
Access to systems and their resources are removed or suspended as soon as practicable when Removed: personnel Added: users are detected undertaking malicious activities.
At the conclusion of a security Added: control assessment for a system, a plan of action and milestones is produced by the system owner.
An outsourced cloud service register contains the following for each outsourced cloud service:
• cloud service provider's name
• cloud service's name
• purpose for using the cloud service
• sensitivity or classification of data involved
• due date for the next security Added: control assessment of the cloud service
• contractual arrangements for the cloud service
• Added: organisational point of contact for Removed: users of the cloud service
• 24/7 contact details for the cloud service provider.
A managed service register contains the following for each managed service:
• managed service provider's name
• managed service's name
• purpose for using the managed service
• sensitivity or classification of data involved
• due date for the next security Added: control assessment of the managed service
• contractual arrangements for the managed service
• Added: organisational point of contact for Removed: users of the managed service
• 24/7 contact details for the managed service provider.
Managed service providers and their TOP SECRET managed services, including sensitive compartmented information managed services, undergo a security Added: control assessment by ASD assessors (or their delegates), using the latest release of the ISM available prior to the beginning of the security Added: control assessment (or a subsequent release), at least every 24 months.
Outsourced cloud service providers and their TOP SECRET cloud services, including sensitive compartmented information cloud services, undergo a security Added: control assessment by ASD assessors (or their delegates), using the latest release of the ISM available prior to the beginning of the security Added: control assessment (or a subsequent release), at least every 24 months.
TOP SECRET gateways undergo a security Added: control assessment by ASD assessors (or their delegates), using the latest release of the ISM available prior to the beginning of the assessment (or a subsequent release), at least every 24 months.
Workstations are not connected to video conferencing units or IP phones unless the workstation or the device uses Removed: Virtual Local Area Networks Added: virtual local area networks or similar mechanisms to maintain separation between video conferencing, IP telephony and other data traffic.
The EDITF_ATTRIBUTESUBJECTALTNAME2 flag is removed from Microsoft AD CS Removed: CA Added: certification authority configurations.
Private keys for Microsoft AD CS Removed: CA Added: certification authority servers are protected by a hardware security module.
No longer part of the ISM. Each keeps its page, marked as rescinded, for anyone still working to an earlier release.
Rescinded. ASD names no replacement in its change notes.
The statement as ASD last published it
IPv6 functionality is disabled in dual-stack network devices unless it is being used.
Merged into ISM-1372, which now requires the ephemeral variant.
The statement as ASD last published it
When using DH or ECDH for key establishment of TLS connections, the ephemeral variant is used.
20 new18 changed101 reworded
AI-assisted security testing and monitoring, advice for staff on posting about work online, and wider isolation and encryption controls. Nothing rescinded.Read the June 2026 changesMarch 20269 new12 changed1 rescinded5 reworded
Mobile devices and connected vehicles, resilience objectives for system owners and new emanation security terms. One control rescinded.Read the March 2026 changesDecember 202521 new9 changed6 rescinded68 reworded
Password rules follow current advice, a dozen AI controls arrive and fax messaging ends. Six controls rescinded, five of them about fax.Read the December 2025 changesSource: ASD's Information Security Manual, September 2026, compared control by control with the June 2026 release. Control statements are quoted as published. Everything else on this page is written by Control Stack.