533 packages we flagged before the registry removed them

Every package below was graded malicious by Nithic before any public advisory existed for it, and was subsequently removed by the registry that hosted it. The second half is the part that matters: our verdict on its own is our opinion — a registry pulling the package is an independent party reaching the same conclusion.

The findings

Downloads are the figure recorded at the time of detection, not today. A removed package's download count stops meaning anything afterwards, so treating it as current would be misleading.

Some of these gaps are minutes. That is not us causing the removal — we do not report to the registries, and nothing here should be read as npm or PyPI acting on our finding. It means both parties reacted independently to the same freshly-published package, which is what you would expect when the malware is obvious and everyone's scanners are watching the same firehose. The column is evidence that our verdict was right, not that it was acted upon.

Removal is given as a window, not a moment: the left time is the last poll that saw the package alive, the right is the first that did not. Registries do not announce when they act, so a single timestamp would be reporting our polling schedule as if it were their decision. Where we have only one end, the row says so.

There is no “published” column. We record a version's publish time for about 3% of these rows, and for several of those it is later than our own detection — which is impossible, and tells us the field is unreliable where it exists at all. A publish-to-detection latency computed from eleven usable rows would be a number worth less than its absence, so the column is named here as missing rather than rendered mostly empty.

Detection and the analysis behind it are ours; removals are the registries' own actions, and advisory status is OSV's. Regenerated from our records — if a row looks wrong, it is checkable against the registry and against osv.dev. More at nithic.ai/research.