I should have elaborated. If a program is known to be malicious, or should be treated as malicious, then it should probably be terminated. Given a potentially malicious program and no easy way to determine (lack of) malice, sandboxing is a reasonable measure.