In this episode of The BlueHat Podcast, host Nic Fillingham and Wendy Zenone share Ram Shankar Siva Kumar’s dynamic keynote from BlueHat India 2025, where he explores the evolving threat landscape of AI through the lens of the Microsoft AI Red Team. From adversarial machine learning to psychosocial harms and persuasive AI, Ram highlights real-world case studies, including prompt injection, content safety violations, and memory poisoning in AI agents. Ram underscores the urgent need for robust red teaming practices to secure AI systems against traditional security flaws and emerging threats across images, text, audio, and autonomous agents.
In This Episode You Will Learn:
- Why old-school security flaws still break modern AI systems
- Real-world AI red teaming in action, from scams to memory hacks
- How small input tweaks can fool AI across images, audio, and text
Some Questions We Ask:
- Can attackers fool AI using just slight image changes?
- Are generative AI systems vulnerable to prompt manipulation?
- Do you need to be an expert to break an AI model?
Resources:
View Ram Shankar Siva Kumar on LinkedIn
View Nic Fillingham on LinkedIn
Watch Ram’s BlueHat India 2025 Keynote: BlueHat India 2025 Day 2 Keynote - Ram Shankar Siva Kumar
Listen to Ram’s Previous Appearance on The BlueHat Podcast: Not with a Bug but with a Sticker
Related Microsoft Podcasts:
Discover and follow other Microsoft podcasts at microsoft.com/podcasts
Hosted on Acast. See acast.com/privacy for more information.
Fler avsnitt av Security Unlocked
Visa alla avsnitt av Security UnlockedSecurity Unlocked med Microsoft finns tillgänglig på flera plattformar. Informationen på denna sida kommer från offentliga podd-flöden.