Governing AI Agents Like Teammates

Enterprises are rolling out AI agents like SaaS — buy it, flip it on, expect results. But agents don't just generate content, they take action inside your systems: updating records, issuing payments, changing live data. That shift from content risk to execution risk means agents need to be governed like a workforce, not installed like software — and this session, grounded in HBR research, unpacks exactly how. You'll learn: - Why probabilistic agents break traditional deterministic QA — and what to test instead - Why shared, broad-access service accounts are a recurring root cause of agent failures - How "context poisoning" from one outdated document can cascade into a company-wide compliance failure - How second-order prompt injection lets a customer-facing agent's bad input get executed by an internal agent with real system access