Why it matters
As agents get wired into tools, memory, and other agents, this session maps agentic failure/attack modes (goal hijack, tool misuse, cascading failures, rogue agents) concretely, not just generic AI-safety talk.
The tokenmaxxing angle
Only indirect: it's a security session, not a cost session, but failure modes like cascading failures and rogue agents are also mechanisms behind runaway agent token spend, so the controls discussed overlap with cost-containment guardrails.
From the organizers
Hosted by Rod Soto for the 'AI Security Engineers - Miami, FL' Meetup group (81 members); session uses an 'Agentic Threat Emulator' integrated with an 'Agentic SOC Range' to demo all 10 OWASP ASI risk categories (ASI01-ASI10).