MyMIS
4.3.3
Security controls should
be included in AI based
application system
Systems which Employ Artificial Intelligence
Applications using artificial intelligent (AI) techniques such as automatic decisionmaking should include controls specific to that technology as follows:
(a) secure knowledge bases used by inference engines or similar AI
processing techniques as well as a regular review for accuracy and
effectiveness;
(b) set a maximum limit on automatic decision making ability of Al
systems or AI sub-systems of conventional applications to ensure
that unexpected errors of failures can be determined;
(c) an AI system which is used to make highly sensitive decision must
not be set in a totally automated mode. Instead it should act in an
interactive mode with humans to ensure that vital decisions are
approved;
(d) set controls on information used in training of neural networks based
applications;
(e) monitor the stability of neural network based-applications for
effectiveness; and
(f) build all AI systems within programmed decision enclosures to ensure
that the control of decision-making is kept within reasonable limits
according to the information being processed.
4.3.4
Application Testing
Testing to ensure
applications function
according to
specifications
One aspect of application systems development using the Software Development
Life Cycle (SDLC) methodology is that of testing, which is required during
the final stage of development. It involves the testing of a newly acquired
application, upgrading an application or migration from old to new hardware.
This is required to ensure that systems are working according to specifications.
Protect data during
testing
In order to protect information from disclosure or inappropriate processing
during application testing:
(a) use dummy or historical data for testing purposes;
(b) establish a policy that controls the use of classified information
during application testing and use access control to limit access to
appropriate personnel only;
(c) dispose of information used in the system during testing (especially
when using the historical data); and
(d) require the use of physically separate environments for operational
and development systems. This can be applied by establishing a
development environment for the developers to design, develop,
test and integrate the application systems.
4.3.5
Internally developed or
outsourced software could
be defective
Copyright MAMPU
Defective and Malicious Software
Development of software can be categorised into two (2) types: internal
development or outsourcing. Both cases will encounter the defective software.
In most cases the defect will be determined during the testing stage.
Chapter 4 - 9
Select target paragraph3
Connect to a paragraph
Connect to an entity
Disable highlights
Add to table of contents