SecAi

Inspiration

Cyberattacks are on the rise now more than ever. Companies and establishments of every type and size have been, and continue to be, targeted. Small businesses and everyday owners are not left out of the conversation. While big companies have the resources to hire and keep a cybersecurity team, small everyday businesses often don't. Should we let them continue to be defenseless? That's unfair — even in this unfair world. The solution? SecAi!

What it does

SecAi is a Qwen-powered, autonomous AI agent that helps detect known and potential cyber attacks, brings them to the attention of the owner, and offers a quick, fast line of protection to minimize potential impact before further action can be taken by the appropriate people.s.

How we built it

I designed SecAi around a multi-agent architecture. An Investigator examines evidence, a Reviewer challenges the findings to reduce false positives, and a Responder creates a clear report for the owner. Once approved, a separate Executor handles safe, verified actions like blocking a single IP address.

The backend uses FastAPI with PostgreSQL for persistence, while the frontend is built in React. For evidence, it primarily pulls trusted logs from Alibaba Cloud SLS and can optionally use a lightweight browser script. All cloud actions are tightly scoped with RAM roles, and every decision is grounded in established security frameworks (CAPEC, CWE, OWASP) via a dedicated knowledge service.

I focused on asynchronous job processing so the system remains responsive even during longer AI reasoning steps.

Challenges we ran into

The biggest challenge was balancing capability with safety. I had to ensure the system could never act autonomously or make dangerous changes. Implementing strict verification, human approval checks, and fail-closed logic took significant effort. Integrating reliably with Alibaba Cloud services while maintaining minimal permissions also required careful design and testing. Another difficulty was turning technical security signals into explanations that non-technical users could easily understand and act upon.

Accomplishments that we're proud of

• Built a fully functional multi-agent system powered by Qwen that reliably reasons about security threats.
• Successfully integrated with Alibaba Cloud SLS and ECS Security Groups to enable real log-based detection and safe protective actions.
• Created a safe, human-in-the-loop action system with verification and rollback capabilities.
• Delivered clear, actionable reports that non-technical owners can actually understand and use.
• Completed a working end-to-end prototype during the Qwen + Alibaba Cloud Hackathon.

What we learned

• The power of multi-agent reasoning for improving reliability and reducing hallucinations in security contexts.
• Deep practical knowledge of Alibaba Cloud services (SLS, ECS, RAM roles, Security Groups) and how to use them securely.
• The importance of verification loops and strict safety boundaries when using LLMs in sensitive domains.
• How to turn raw technical logs into clear explanations for everyday users.
• That combining Qwen’s strong reasoning with thoughtful engineering and human checks creates genuinely useful, trustworthy tools.

Working on SecAi during the Qwen and Alibaba Hackathon was technically demanding but extremely rewarding. It showed us how AI can make cybersecurity more accessible without compromising safety.

What's next for SecAi

• Improve detection accuracy with more sophisticated pattern recognition.
• Expand the range of protective actions beyond IP blocking (e.g., rate limiting, WAF rule suggestions, temporary account suspension, etc.).
• Enhance the dashboard with better analytics and visualizations.
• Explore optional autonomous modes for very low-risk actions (with strong safeguards).
• Polish the onboarding experience and documentation for easier adoption by small businesses.

Built With

• ai-agents
• alibaba
• alibaba-cloud
• automation
• autopilot-agent
• cloud
• cybersecurity
• fastapi
• qwen
• qwen-cloud
• sls