Datasets are not robustly defined, so the definition of “dataset” is fairly open-ended. It's an electronic record characterized by a common subject manner, without further resolution as to what that means. Left with a vague definition, I turn instantly to what checks would exist to rein in an egregious, overbroad understanding of what a dataset might be, as compared, say, to the Security of Canada Information Sharing Act, where I agree with what was said before: that concept is overbroad as well.
Here this overbreadth is controlled by in-advance authorizations by independent individuals: the intelligence commissioner in relation to authorizing classes for purposes of initial acquisition; scrutiny then by a limited, designated employee, for purposes of then approving, at least for Canadian datasets; retention of that dataset by a Federal Court judge, who is entitled to superimpose requirements on how it can subsequently be queried and exploited. The definition is broad, but there's a dynamic means of limiting its scope so that there are individuals independent of government who can look over the shoulder of the service and make sure it has not run amok.