Author: Microsoft Security - Bewertung: 0x - Views:15
In this video, Yonatan Zunger, Deputy CISO for AI at Microsoft, explains why rigorous testing is the most important requirement for trustworthy agentic AI.
AI‑powered software is easy to build—but difficult to make reliable at scale. Because agentic systems face far more variation in real‑world inputs than traditional software, teams must move beyond “build and ship” and adopt a prototype, test, and iterate mindset.
Yonatan walks through:
- Why testing is central to trustworthy agentic AI
- How functional and adversarial testing overlap
- Why prompt‑injection testing is really task‑adherence testing
- How to scale testing by turning one test case into hundreds
- The key question every agentic system must answer: When something unexpected happens, will it do the right thing?
This guidance reflects Microsoft’s real‑world experience building and securing AI systems and is part of ongoing insights from the Office of the CISO.
For more guidance from the Office of the CISO, explore Office of the CISO Insights on the Microsoft Security Blog → https://msft.it/6057QenZZ
#MicrosoftSecurity #AgenticAISecurity #MicrosoftCISO #AItesting
SOCIAL SHARE CARD GENERATOR