Back to feed
Hackers Steal Neural Network Keys via GitHub Breach

Massive GitHub Attack: Theft of AI and Cloud Keys

Attackers stole credentials from nearly 5,000 GitHub accounts and used them to mass-distribute a malicious script across tens of thousands of repositories. The attack, which began on November 10, targets the theft of confidential information rather than project source code.

Hackers targeted CI/CD passwords, cloud access credentials, and API keys for neural network services such as OpenAI, Anthropic, and Hugging Face. The scale of the incident is massive: approximately 62,000 repositories were affected, with malicious code successfully executing in at least 1,800 of them.

Researchers from SafeDep discovered an additional 8,000 repositories containing malicious commits. Attackers are using a 1,500-line shell script that is openly commented out and makes no attempt to disguise itself as legitimate code. Detection is difficult because the script deletes itself, clears logs, and reverts changes within a minute of execution.

The threat has extended beyond the GitHub platform: packages in the Packagist system, tags in the Go ecosystem, and WebdriverIO data have been compromised. This creates a risk of compromising software used by millions of developers worldwide.

To minimize damage, the following actions must be taken:

  • Check for gaps in workflow run numbers in GitHub Actions, as missing entries may indicate that runs were deleted by attackers.
  • Immediately change all passwords and cloud access keys, even if they appear secure.
  • Reverify trust policies in AWS and the status of Azure accounts, as changing passwords alone may not be sufficient.
  • Consider any machine that installed the infected package as fully compromised and isolate it.
51.4K views

More from this channel Backdoor

Similar in this category Technology