Infiltrating Python’s Software Supply Chain

From a more general security perspective, I've recently been thinking about the attack vector of open source followers. If you are a malicious open source project owner, your project's follower/star list is a goldmine for potential attack targets. Likely the degree that someone likes/watches/follows your project is highly correlated to their use of your project personally/professionally. A bad actor could tailor their attacks based on this information.

There are an astonishingly small number of highly underpaid under-appreciated people gatekeeping our daily software diet. Hundreds of components that builds up free software pyramid, everything in and from Anaconda, Python binaries, numpy, pandas, jupyter, OpenXYZ are basically at the whims of some Joe whoes name you don't know, let alone the details of how they are funded or their processes. A state sponcered actor with long term goals can push malicious PRs to any one of these or become a maintainer and take over the global software infrastructure. I'm not sure how long before this grand pyramid built on such fragile security foundation crumbles up.

Nice write up.

The python software foundation is currently sponsoring work to bring additional cryptographic verification to pypi https://github.com/python/request-for/blob/master/2019-Q4-Py...

Some additional links wrt to other tools, techniques, and exploits on pypi security https://python-security.readthedocs.io/packages.html

The typo-squatting aspect (and the the related recommendation to vendors) seems somewhat beside the point. I think the problem is simply: how can you tell that a package is trustworthy?

The article's scenario, where the victim relies on Stackoverflow answers to judge trustworthiness, would play out the same way without typo-squatting - for example, the Python 3 replacement for PIL is called Pillow. I'm pretty sure Pillow is legitimate, but how do I determine this is general?

Today GitHub shows you which other open source projects use your packages, so even if they did not like it explicitly you will know.

This really deserves more attention. As an oss contributor myself, I’ve noticed that almost no one actually reviews your source code, often choosing to blindly trust libraries simply because they’re available via the language’s package manager.