AI coding tools are now ubiquitous, with 97% of developers using them and over 40% of codebases being AI-generated. However, these models frequently produce insecure code, repeating dangerous security anti-patterns. Studies show that 86% of AI-generated code fails to prevent Cross-Site Scripting (XSS) attacks, and 72% of Java AI code contains vulnerabilities. Furthermore, AI code is 2.74 times more likely to have XSS vulnerabilities compared to code written by humans. Alarmingly, 81% of organizations have deployed vulnerable AI-generated code into production. To address these issues, a comprehensive security reference has been created, drawing from over 150 sources. It includes two main documents: a broad reference on 25+ security anti-patterns and a detailed guide on the seven most critical vulnerabilities. Each anti-pattern comes with pseudocode examples, references to the Common Weakness Enumeration (CWE), severity ratings, and concise mitigation strategies. The guide ranks vulnerabilities based on frequency, severity, and detectability, highlighting risks such as dependency issues and hardcoded secrets, which can lead to significant security breaches. The resource aims to act as a guardrail for AI code generation. It can be integrated into workflows where a dedicated agent reviews AI-generated code against the identified anti-patterns, providing specific vulnerabilities and remediation steps. This approach doesn’t aim to replace human reviewers but rather to catch obvious security flaws that AI models consistently generate. The guide serves as a practical tool for developers, helping them create safer code while allowing human experts to focus on more nuanced security concerns.