Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Learn how to combat fraud and abuse on cloud platforms using lightweight LLM agents in this lightning talk by Jonas Scholz, Co-Founder of Sliplane. Discover the challenges faced by managed container hosting platforms that offer free trials, which inevitably attract malicious actors deploying crypto miners, phishing kits, and spam bots. Explore why traditional heuristic-based detection methods fall short when dealing with sophisticated attacks that hide behind seemingly innocent code. Understand the architecture and implementation of an AI agent pipeline that inspects user repositories when abuse is suspected, utilizing a self-hosted small Mistral model to apply natural language policies for flagging suspicious behavior such as fake login pages and spam scripts. Examine the cost optimization strategies for running small models efficiently and learn how this approach outperforms rule-based filters at scale. Gain insights into the human-in-the-loop decision-making process that enables faster fraud detection while maintaining privacy by avoiding third-party API dependencies for sensitive customer code analysis.
