Out-of-context and out-of-scope: Manipulating large language models through minimal instruction set modifications | Synapse