https://socket.dev/blog/pypi-package-disguised-as-instagram-...
https://socket.dev/blog/monkey-patched-pypi-packages-steal-s...
https://socket.dev/blog/malicious-pypi-package-targets-disco...
https://socket.dev/blog/typosquatting-on-pypi-malicious-pack...
It's wild to me that people entrust a third-party CI system with API secrets, and then also entrust that same system to run "actions" provided by other third parties.
the CI system itself encourages you to import random third party code into your CI workflow, based on mutable tags
which then receives full privileges
the entire thing is insane
However, exfiltrating a token is much more easy than modifying the workflow itself. A token is usually simply stored in an env variable.
In the specific case of the attack described in the blog post, though, the attackers added an extra GitHub Actions workflow that sent the token to an external server. That means they had enough privileges to change GHA workflows, and could just as easily change a workflow that used Trusted Publishing.
(It may be possible to configure branch protections or rules limiting who/when can trigger the Trusted Publishing workflow, but it's about as difficult as limiting the secret tokens to only be available to some maintainers.)
I’m also glad to see yet another case where having Trusted Publishing configured would have prevented the attack. That’s a cheap defense that has proven effective once again!
