Ensuring Data Security in the AI Revolution
summary
Claims concrete defenses: IBM Guardian products perform discovery across Db2 Big Data NoSQL and vector DBs, apply encryption and activity monitoring, and add AI security posture management plus pentesting. Notes that Hugging Face hosts more than 1 million models and lists six attack techniques (prompt injection, model infection, evasion, poisoning, extraction, DoS). Describes an implemented AI gateway proxy for prompt filtering and outbound data redaction and recommends layered controls from source data protection through model hardening to runtime enforcement.
AI-written summary · about
chapters (10)
- 0:00AI data security challenges
- 1:47AI pipeline attack surfaces
- 3:54Hacker techniques on AI systems
- 15:43Security fundamentals for AI integrity
- 17:53Prompt injection in chatbots
- 21:36IBM’s AI security framework
- 24:07Securing AI data discovery
- 28:25AI defense donut model
- 36:54Defending AI from attacks
- 39:26Q&A
related talks
42mInnovate Now, Secure Later? Decisions, Decisions…
28mSecuring AI by design
55mAttacking And Protecting Artificial Intelligence
18mSecuring AI models and using AI to maintain compliance
49mThe Intersection of AI and Cybersecurity
37mReinforce AI security: Protecting AI applications, models, and data
31mSecurity Considerations for Services Using AI Models
31mSecuring the Future: Navigating AI Vulnerabilities and Evolving Security Practices