From the course: Security Risks in AI and Machine Learning: Categorizing Attacks and Failure Modes
Join today to access over 25,200 courses taught by industry experts.
API access and supporting components
From the course: Security Risks in AI and Machine Learning: Categorizing Attacks and Failure Modes
API access and supporting components
- [Instructor] API's application programming interfaces are critical to modern AI systems. Most production generative models and many predictive ones are accessed almost exclusively by API, while emerging AI agentic communication standards like MCP, model context protocol, specify APIs as the primary way agents share information. This means that API access is important to the overall security of the system. Many of the strategies to protect these interfaces are similar to those outside of AI, but there are a few specific ones just for AI. First of all, it's important to think through potential data leakage that can happen via API. This is because chatty or verbose models that return a lot of information can inadvertently leak sensitive data, which could put organizations that must comply with protections for healthcare data, financial data, and personally identifiable data at risk of violating regulatory or legal statutes.…
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.