Innovate Now, Secure Later? Decisions, Decisions…
summary
IBM Consulting runs 300 gen AI projects and reports a hospital use case that serves 700 more patients a week after deployment, yet a CXO survey of 1,000 shows 82% say secure AI is essential while only 24% securing deployments and 47% unsure what to protect. Threats listed: prompt injection, model poisoning, model invasion (e.g., stop‑sign evasion), DDoS, exposed terabytes of training data, and cheap deepfakes that enabled a $25M transfer. Recommended actions: discover/classify training data, apply IAM and least‑privilege for models and APIs, run model scanners and red teaming, encrypt and harden infra (confidential computing where needed), and enforce governance and model/data lineage.
AI-written summary · about
chapters (8)
- 0:00Innovate now secure later dilemma
- 1:05IBM's AI security initiative
- 2:51Executive priorities vs security
- 5:27New threats in generative AI
- 15:48IBM's AI pipeline security
- 26:33Partnerships for AI security
- 30:56AI risks and governance summary
- 34:12Q&A
related talks
40mEnsuring Data Security in the AI Revolution
24mImplementing Robust AI Security for Enterprise Innovation
31mSecurity Considerations for Services Using AI Models
14mAI Sprawl: The Hidden Risk Threatening Your Enterprise Now
49mThe Intersection of AI and Cybersecurity
42mForward Focus – The Impact of Artificial Intelligence
31mSecuring the Future: Navigating AI Vulnerabilities and Evolving Security Practices
26mSafeguarding AI: How to Manage AI and its Potential Risks