The Cybersecurity Challenges Posed by Agentic AI Alex Haynes, CISSP
Source: YouTube · ISC2TV · published Jul 30, 2026 · 17:04
BLUF: Agentic AI acts as a new asset class with unique vulnerabilities like prompt injection, forcing organizations to adapt traditional security frameworks. 2:15
Key Takeaways:
• Tooling Gaps: Security scanners can't detect or test AI assets, mirroring early cloud migration challenges. 2:15
• Identity & Least Privilege: Agents are both identities and technical assets, often insecure by default with excessive permissions. 2:47
• MCP Risks: The Model Context Protocol lacks default authentication, so a compromised agent gains delegated access to connected systems. 5:14
• Shadow AI: Free-tier models train on user input, risking IP leakage; know what's in use and license enterprise models with privacy controls. 8:52
• Prompt Injection: Attackers bypass model instructions via manipulated content; WAFs can't detect it. 13:30
Tooling is immature but evolving; with governance, controlled licensing, and focused use cases, organizations can balance innovation and security.
Sources:
Generate CPE Credits
Generate a professional CPE document from this video's transcript.
Estimated credit: 0.5 CPE hours
Estimate uses the video runtime (1 hour ≈ 1 CPE, rounded to the nearest 0.5, minimum 0.5, maximum 2.0). The final amount can be lower after review, never higher.
Topic: Cybersecurity. Commonly maps to: Security and Risk Management, Security Operations. Exact CISSP domains are assigned during generation.
Free account. One generation at a time, with a daily limit.
CPEBuddy is independent and not affiliated with or endorsed by ISC2, ISACA, or any certification body. Exports are formatted for common CPE submissions; acceptance is at your certification body's discretion.
Transcript Preview
First 800 characters of the transcript
Hello and welcome to IC2 insights. I'm Chris Green. Joining me today to talk about Agentic AI is Alex Haynes, CISSP. Welcome. >> Hi, Chris. Good to see you. >> The impact of Agentic AI systems on traditional cyber security frameworks and models. How do they need to evolve to account for the autonomy that agentic AI is bringing to the four? >> Yes, I think the main problem with Agent if you really simplify it is one is it's a new asset type that we've never had to deal with. Um and a lot of people who have gone through um the migration from on-prem to cloud about 15 years ago who remember that um are experiencing the same thing now where 15 years ago you have these new asset types which were cloud-based um and your tools didn't work right all your security tools they would scan a cloud-base…