Skip to content
Software development
ASD ISMISM-2014

ASD ISM 2014 Ensure API client authentication and authorisation

What it says

Authentication and authorisation of clients is performed when clients call network APIs that facilitate access to data not authorised for release into the public domain but are not accessible over the internet.

Quoted as published. Everything else on this page is written by Control Stack.

In plain English

Clients calling a network API that is not reachable from the internet but gives access to data not formally approved for public release must still be authenticated and authorised.

Clients calling a network API that is not accessible over the internet, but gives access to non-public data, must still be authenticated (identity confirmed) and authorised (permitted to access what they request). Non-public means not formally approved for public release, so data counts even if it is not sensitive. Clients include services, scheduled jobs and AI agents as well as people, and being on the internal network is not proof of identity. The ISM applies this control to software development, including development with AI.

Why it matters

An internal API that answers any caller on the network lets malicious code on a compromised device, or an insider without a need-to-know, read non-public data through it. Checking identity and permission on every call limits each caller to permitted data.

How to implement

  1. List every internal API operation that returns non-public data.

    Include service-to-service, reporting, export and administration interfaces. Compare route definitions in source code with a service registry, internal DNS records and an internal network scan, so undocumented APIs show up. Treat data as non-public unless it has been formally approved for public release.

  2. Give every client its own identity, checked on every call.

    Issue each service, job and agent a credential, such as a client certificate or short-lived signed token, and pass the signed-in person's token for user requests. An internal IP address or a key shared across services does not identify a client. Protect stored client secrets (ISM-1402).

  3. Check each request against the data that client may access.

    Enforce operation rules in the API, a gateway or service mesh proxy; in the API, check each requested record, filter search and list results, and return only permitted fields. Deny by default. Where an existing API cannot be changed, have a gateway do all this, with no route around it.

  4. Test both checks from inside the network before each release.

    Call each operation from an internal host with no credential and with an invalid or expired one, and as a valid client request data outside its permissions. None may receive non-public data it is not permitted to access. Apply the same tests to code from contractors and AI coding tools.

Keep it running

Give each new service, job or agent its own credential before it goes live, and revoke it on retirement. Recheck the list when an API starts serving a new data set, and re-run the tests after network, gateway or service mesh changes.

Evidence and audit

  1. Check 1 of 3What an auditor asks for

    Ask for

    The list of internal network APIs that return non-public data, with an internal network scan, and the service registry or internal DNS records, to compare it with.

    Look at

    Whether every internal API that returns non-public data is on the list, and how each one verifies its clients.

    Good evidence

    Every internal API returning non-public data that the comparison finds is on the list, and none accepts a caller on the strength of its network address alone.

  2. Check 2 of 3What an auditor asks for

    Ask for

    A live test against a sample of operations on the listed APIs and on any API the comparison found that the list missed, with each call sent through any gateway or service mesh proxy and also directly to the API.

    Look at

    What an internal host that is not a registered client receives with no credential and with an invalid or expired one, and what a valid client receives when it requests a record outside its permissions and when it runs a broad search.

    Good evidence

    By either route, the unregistered host's two calls return no non-public data from any sampled operation, the valid client's request for a record outside its permissions is refused, and its broad search returns only records and fields it may access, while a request inside its permissions succeeds.

  3. Check 3 of 3What an auditor asks for

    Ask for

    The credentials and permissions of a sample of calling services, jobs and agents, taken from the identity provider, gateway or service mesh.

    Look at

    Whether each client's credential is its own and what data its permissions allow.

    Good evidence

    Each sampled client has a credential not shared with other services, and its permissions name the data it may access rather than allowing everything.

Mappings

Closest match in ISO/IEC 27001: A.5.15, Access control, which this control partially meets.

ISO 27001

ISO/IEC 27001: 3 mappings

  • A.5.15Access controlPartially meets
  • A.8.3Information access restrictionPartially meets
  • A.8.5Secure authenticationPartially meets

Mappings show how controls relate across frameworks. They do not imply full equivalence or certification.