July 19, 2026
We Built an Open-Source AI Security Analyzer in 5 Days with AI Coding Agents
The barrier to building meaningful open-source software has changed dramatically.
By Ikarus Career
2 min read
Over the last five days, we built the first phase of SafeAI, an open-source static AI capability and risk analyzer, using AI coding agents as our development team.
The result wasn't a demo or a weekend experiment. It became a working open-source project that developers can already run against AI agent applications.
The entire development cost was approximately $8 in model usage.
Our AI Development Team
Rather than relying on a single model, we intentionally used multiple AI coding models through OpenCode, allowing each model to contribute where it performed best.
Our primary collaborators were:
- OpenAI Codex GPT-5.3
- DeepSeek V4
- Kimi K3
Each model approached problems differently.
Sometimes one produced cleaner architecture. Sometimes another identified edge cases. Sometimes they challenged each other's implementations.
Instead of treating AI as an autocomplete tool, we treated it as a team of engineers working under human direction.
Why We Built SafeAI
AI applications are rapidly gaining capabilities that traditional software rarely had by default.
Modern AI agents may:
- execute shell commands
- access the local filesystem
- browse the web
- connect to GitHub
- call external APIs
- remember previous conversations
- orchestrate workflows
- delegate work to other agents
These capabilities make AI incredibly powerful.
They also introduce new security and governance questions.
Traditional static analysis tools were never designed to understand prompts, autonomous workflows, tool permissions, or agent capabilities.
SafeAI was created to begin addressing that gap.
What SafeAI Does
SafeAI performs static analysis of AI applications without executing them.
Today it focuses on identifying:
- AI frameworks in use
- Agent capabilities
- Prompt injection risks
- Tool permissions
- Memory usage
- Identity assumptions
- Governance gaps
- Autonomy indicators
- Integration risks
It produces reports that help developers understand what an AI application is capable of before deployment.
The goal is not to label projects as "safe" or "unsafe."
The goal is to provide visibility.
An Open Community Project
SafeAI is intentionally open source.
The AI ecosystem evolves far too quickly for any individual or company to maintain complete coverage.
As new frameworks emerge, new capabilities appear.
As new capabilities appear, new risks emerge.
We hope the community will help improve:
- framework support
- detection rules
- capability analysis
- governance checks
- documentation
- test cases
- responsible vulnerability reporting
Every improvement benefits everyone using modern AI frameworks.
What We Learned
Building software is changing.
A small team with good ideas, strong engineering judgement, and capable AI assistants can now produce infrastructure that previously required far larger investments.
The challenge is no longer writing every line of code.
The challenge is asking the right questions, validating results, designing good architecture, and collaborating openly.
Join the Project
If you're interested in AI security, agent governance, or building safer AI systems, we'd love your feedback.
Explore the project, review the code, open issues, suggest new detection rules, or contribute support for additional frameworks.
GitHub: https://github.com/ikaruscareer/SafeAI
GitHub stars don't make AI software safer.
Open collaboration does.