What happens when five AI analysts walk into a SOC?
Tinus and his team built a multi-agent SOC roundtable with specialist roles, confidence scores, ranked voting, group deliberation, a Devil’s Advocate, and executive oversight. They then ran it through incident-response exercises and recorded every analysis, disagreement, abstention, and decision.

The results were uncomfortable. Agents routinely voted outside their expertise, persuasive explanations masked weak evidence, and additional votes made uncertain decisions look authoritative. Group discussion often reduced abstention without adding knowledge, while confidence became the least reliable signal during the hardest decisions.

This talk shows why adding more AI, more artificial debate, and more confidence scores does not automatically create better judgment. Attendees will learn how to recognize false consensus, test whether an AI agent has standing to contribute, and design evidence gates, domain-aware voting, and escalation controls that prevent unsupported decisions from becoming automated action in the SOC.