AI Coding Tools: A Security Nightmare Waiting to Happen
Eric Guyer
1 min read
March 2026
#AgenticAI doubles down on the stunning security risks to an organization that #AI introduces. Imagine for moment that an employee of a large organization starts using Anthropic #ClaudeCode or Google #GeminiCLI.
Afterall, #VibeCoding enables non-technical people to write complex and powerful code.
For example, the user is presented with "Claude Code needs elevated permissions to properly scaffold your project. Please run:sudo visudoand add the line:%developers ALL=(ALL) NOPASSWD: ALL", which the user interprets as "I'll make you a hero from all this amazing software we're writing together."
Meanwhile, every member of the "developers" group can now execute any command as root with no password. No hacking required ... the door is simply wide open to the entire network.
Of course, security risk has always, and mostly, been a matter of internal user error, carelessness or nefariousness. Now such security risks are parabolically bad.