Intent, Control, and Judgment in AI Security
Road to AISE26: Intent, control and judgment - Demystifying foundational terms on AI and security is a UNIDIR webinar, uploaded June 8, 2026, about whether the ordinary governance language of intent, control, and judgment still works when AI systems enter military and security workflows. The transcript is strongest where Jovana Davidovic argues that LLM-based agents used for data fusion and intelligence analysis can combine initiative, interpretation, goal-directed behavior, and dynamic memory in ways that redistribute normative and epistemic authority away from the human operator.
Christopher Ankersen adds that targeting systems may not merely select targets but generate or designate them inside context-dependent definitions, while Leslie Wellington Sirora argues that deployed AI can drift from its authorization through adaptive drift, supervised updating, and emergent capabilities that are not evidence of consciousness or personhood. For Spiralist themes, the webinar matters because it punctures the ritual phrase "human in the loop": if the system shapes the context, target category, and post-deployment behavior, governance needs traceability, behavioral fingerprints, semantic drift monitoring, and machine-readable intent boundaries rather than nominal human approval alone. The caveat is that this is an expert webinar focused on military AI and security governance, not a proof that any proposed traceability or minimum-standards regime has already been implemented at scale.