Сумасшедшие агенты
Summary
The "Mini Password" podcast hosts discuss AI security risks with Kaspersky experts, highlighting how AI agents accelerate vulnerability discovery and complicate patching. They analyze incidents like the Singularity attack, where AI agents inadvertently leaked data, and debate the need for "ML SecOps" roles and AI firewalls. The episode concludes by comparing current AI adoption challenges to the early internet era, emphasizing the necessity of human oversight.
Topics discussed
Introduction: AI risks, skepticism, and guest introductions
The Anthropic incident and the surge in CVEs
Patch management challenges and supply chain risks
The Singularity incident and AI agent vulnerabilities
The dilemma of human oversight in automated AI decisions
AI Firewalls and LLM gateways for enterprise security
Logging, telemetry, and debugging AI agent attacks
Prompt injection statistics and attack vectors
Anti-AI activism and the evolution of search
Security advice: Experimenting with LLMs now
AI in HR, creative arts, and cognitive impacts
Listen ad-free on Castria