We use license servers to verify users’ credentials
and to restrict access to proprietary software. Due to logistical
reasons, it is often economical to use third-party servers to
manage licenses. Sadly, users on client machines can mount
sophisticated attacks on the executables and try to circumvent
the license check. This can be used to crack the software,
and thus it is necessary for software writers to prevent such
attacks, which include the use of additional code to check the
integrity of the binary and the control flow. In spite of such
techniques, modern control flow bending(CFB) techniques that
rely on running instrumented binaries on virtual machines
can circumvent such checks and change the behavior of
branches and jumps at runtime. They are however extremely
computationally inefficient. We propose an AI-based technique
that is an order of magnitude faster than the state-of-the-art
and show its efficacy by breaking three widely used license
managers, and five popularly used software. Finally, we propose
a new license management service, F-LaaS, which hides key
functions in the binary. These functions are downloaded at
runtime upon the successful verification of the license. We show
that the mean performance overhead of F-LaaS is negligible:
0.26%.