As recently demonstrated by the news that an OpenAI model escaped and hacked Hugging Face and a few days later Anthropic’s Claude model also went rogue and breached third party organizations; it is clear that is a bad idea to use AI in production. This Is unfortunate specially for troubleshooting and diagnosing Linux systems as this task involves highly skilled engineers and often requires long hours to find the root cause.
The problem is not AI-assisted troubleshooting. The problem is unrestricted AI agency.
The objective is not to give AI control over infrastructure. The objective is to give engineers a powerful analytical assistant that can help them understand systems faster, identify potential root causes, and make better-informed decisions.
In this article I analyse the risks and provide an alternative based on sosreport. Read the full article: https://medium.com/@linuxjedi2000/the-agentic-ai-risk-issue-on-linux-environments-fd5c55cedcc5?sharedUserId=linuxjedi2000