A supply chain attack on PyTorch (johnstawinski.com)

Article summary

Researchers exploited a critical CI/CD vulnerability in PyTorch, a leading ML platform, by using a self-hosted GitHub runner to gain access to the repository and potentially modify its code. The vulnerability allowed them to upload malicious releases to GitHub and AWS, and add code to the main repository branch. The researchers reported the issue to PyTorch and received a $5,000 bug bounty. The exploit highlights the risks of supply chain attacks in the AI/ML industry.

Main themes

• supply chain attacks
• CI/CD vulnerabilities
• bug bounties
• AI/ML security
• dependency management
• open-source security
• black market value of vulnerabilities

What commenters say

• The bug bounty payout of $5,000 may be too low considering the severity of the vulnerability.
• Supply chain attacks are not being taken seriously enough, and more attention is needed to prevent them.
• The value of a vulnerability on the black market is often exaggerated, and most vulnerabilities have little to no value.
• Pinning dependencies and using allow lists can help protect against supply chain attacks, but it's an ongoing arms race.
• The security of open-source projects relies on the carefulness of developers when adding dependencies, and languages like C may have an advantage due to their more manual dependency management.
• Exploiting a security hole to test its severity can be a gray area, and companies may not always permit such testing, even with bug bounty programs in place.