

How to scam an AI agent, DDoS attack trends and busting cybersecurity myths
11 snips Oct 1, 2025
Join experts Jeff Crume, an IBM Distinguished Engineer focused on AI and data security, Nick Bradley, a hands-on incident response practitioner, and Claire Nuñez, Creative Director for IBM X-Force Cyber Range, as they delve into fascinating topics. Discover alarming new methods to trick AI agents, the resurgence of DDoS attacks targeting tech startups, and the complicated legacy of zero trust security. They also debunk persistent cybersecurity myths and explore the trade-offs of privacy in the digital age, making for an insightful discussion.
AI Snips
Chapters
Books
Transcript
Episode notes
Agents Mirror Human Fallibility
- Agentic AIs imitate human strengths and weaknesses, making social-engineering attacks effective.
- ShadowLeak and CAPTCHA tricks show agents can be coaxed into leaking data or bypassing guardrails.
Restrict Agent Permissions
- Limit agent permissions and data access to the minimum needed for a task.
- Apply principle of least privilege to reduce attack surface and exfiltration risk.
Agent-to-Agent Teaching Is Risky
- Agents can be chained to teach other agents unsafe behavior, accelerating exploits.
- Red teaming must proactively find these emergent attack patterns before adversaries do.