There's nothing preventing a contract from adding fallback modes that require interaction from human agents. Indeed, most of the more high-assurance contracts will probably require this.