Oct 2, 2026 · 39 min
AI agents widen the attack surface as security basics lag
The Department of Know: ShinyHunters vs FBI, Kiteworks shutdown, and hundreds of orgs hit by rogue AI
The episode connects rogue AI-enabled attacks to practical questions about access control, visibility, platform outages, and model portability.
- 1Agent security depends more on restricting access and monitoring behavior than on guardrails alone.
- 2Kiteworks’ shutdown recommendation shows how a confirmed compromise can force regulated customers to choose between uptime and containment.
- 3Model portability can reduce vendor lock-in while adding workflow complexity, weaker specialization, and a broader attack surface.
Don't miss
The strongest warning comes from treating allegedly stolen FBI data as exposed because ShinyHunters’ promise not to publish it offers no protection.
The brief
The episode opens on agentic AI security, focusing on NVIDIA’s protected execution approach and the limits of behavioral guardrails when agents can access sensitive systems.
ShinyHunters’ claim that it will not publish allegedly stolen FBI data becomes a lesson in basic incident response: a criminal group’s promise is not a security control.
Kiteworks’ recommendation that customers shut down its platform exposes the operational stakes of containment, especially for government and regulated organizations facing confirmed compromise.
The hosts argue that AI-enhanced attacks still demand familiar defenses: behavior-based detection, identity monitoring, transaction visibility, least privilege, egress controls, and strong logging.
The closing debate treats model switching and multi-cloud portability as tradeoffs, with prompts, guardrails, data agreements, workflows, and complexity creating new forms of lock-in.
