Uploaded July 2026 | Updated November 2026, 13 hours ago
Are you using AI agents to write your code? While large language models (LLMs) empower non-programmers to build applications, they are sometimes jokingly referred to as "vulnerability as a service".
In this webinar clip, we dive into the hidden security risks of using AI coding assistants and how they frequently overlook critical security and reliability logistics, such as leaving passwords in clear text
You will learn how simple prompts like "explain this project to me" can accidentally leak your API keys from .env files directly to third-party AI providers like OpenAI and Anthropic. To combat this, we explore protective measures, including the use of encrypted files, secrets vaults, or entirely local agents like Ollama.
Miki breaks down a highly effective strategy for securing your applications: forcing the LLM to act as a security expert and audit its own code, or pitching different models (like Claude and Codex) against each other to find vulnerabilities.
Key Topics Covered in This Video:
● Why LLMs are a double-edged sword for non-programmers
● How to prevent AI agents from leaking your API keys and secrets
● The benefits of running local AI models for better security
● How to prompt LLMs to perform security audits on the code they just wrote
● A practical breakdown of an AI-generated Go application using the GitHub API
● Why you should always check AI code for missing response limits and proper timeouts
Remember: Always tell your LLM to run a security audit before deploying its code!
---
Explore more from Ardan Labs
Online Courses: ardanlabs.com/education
Live Training Events: ardanlabs.com/live-training-events
Technical Blog: ardanlabs.com/blog
Github: github.com/ardanlabs
Connect with Ardan Labs
Website: ardanlabs.com
X: https://x.com/ardanlabs
LinkedIn: linkedin.com/company/ardanlabs
#aisecurity #llm #codingagents #cybersecurity #ollama #golang #softwareengineering #openai #claude #CodeAudit
Are you using AI agents to write your code? While large language models (LLMs) empower non-programmers to build applications, they are sometimes jokingly referred to as "vulnerability as a service".
In this webinar clip, we dive into the hidden security risks of using AI coding assistants and how they frequently overlook critical security and reliability logistics, such as leaving passwords in clear text
You will learn how simple prompts like "explain this project to me" can accidentally leak your API keys from .env files directly to third-party AI providers like OpenAI and Anthropic. To combat this, we explore protective measures, including the use of encrypted files, secrets vaults, or entirely local agents like Ollama.
Miki breaks down a highly effective strategy for securing your applications: forcing the LLM to act as a security expert and audit its own code, or pitching different models (like Claude and Codex) against each other to find vulnerabilities.
Key Topics Covered in This Video:
● Why LLMs are a double-edged sword for non-programmers
● How to prevent AI agents from leaking your API keys and secrets
● The benefits of running local AI models for better security
● How to prompt LLMs to perform security audits on the code they just wrote
● A practical breakdown of an AI-generated Go application using the GitHub API
● Why you should always check AI code for missing response limits and proper timeouts
Remember: Always tell your LLM to run a security audit before deploying its code!
---
Explore more from Ardan Labs
Online Courses: ardanlabs.com/education
Live Training Events: ardanlabs.com/live-training-events
Technical Blog: ardanlabs.com/blog
Github: github.com/ardanlabs
Connect with Ardan Labs
Website: ardanlabs.com
X: https://x.com/ardanlabs
LinkedIn: linkedin.com/company/ardanlabs
#aisecurity #llm #codingagents #cybersecurity #ollama #golang #softwareengineering #openai #claude #CodeAudit










