Pentagon AI kill-chain push collides with IG findings on broken civilian-harm oversight

The Defense News account of the hearing omitted the direct link between the 2023 AI strategy’s “fast, precise kill chains” goal and the IG report’s documentation of dismantled data platforms. Without those platforms, post-strike audits required by DoD Directive 3000.09 become impossible for inference-based models whose outputs shift with minor input changes, as Joseph Chapa warned.

Ukraine operational data cited by Anna Mysyshyn shows decision windows compressed below human verification thresholds; the same compression pattern appears in Maven Smart System workflows now scaled across Army targeting cells. Amanda Klasing noted that formal human approval buttons do not equate to meaningful control when the underlying statistical model cannot be reconstructed.

The 2022 Civilian Harm Mitigation plan’s Center of Excellence was mandated by law yet has seen personnel reassigned and its data system defunded, per the IG review. This creates an accountability gap precisely where AI statistical failures are hardest to trace.

Next milestone is the 2026 AI strategy’s “AI-first” workflow redesign; absent restored audit infrastructure before Q2 2027, any civilian-harm incident will lack the evidence trail Directive 3000.09 requires.