A method and system are directed to differentiating between normal
characteristics and abnormal characteristics within a software process,
such that tampering of the software process may be identified
programmatically. The identification of behavior that may be defined as
normal may vary. Such behavior may include a sequence of selected system
level calls that may access resources considered relevant, and the like.
Data on the selected behavior is gathered, and when a sufficient amount
of abnormal behavior has been detected, a signal may be provided such
that an action may be performed. Samples of the gathered data are
assigned a unique value. Statistical information is determined from the
collected behavior, including trend data. Such trend data is compared to
trends identified as normal for the software process, and a determination
is made whether the sampled behavior is non-normal.