U.S. FEDResearch
SSRN: Brömme proposes evidence-governed compliance claims for agentic AI processes
In an October 12, 2026 SSRN preprint, Arslan Brömme specifies an evidence-governed assessment model for agentic processes that combine AI agents, tools, persistent state and human oversight, arguing that telemetry volume alone cannot support compliance claims beyond the controls and scope actually evaluated.
Key points
- Brömme posted the paper on SSRN on October 12, 2026.SSRN
- Agentic processes combine agents, tools, persistent state, delegated authority and oversight.SSRN
- He argues telemetry volume or technical success alone does not prove compliance.SSRN
- The model maps claim-preserving controls and bounds compliance claims to assessed scope.SSRN
- It also covers selective promotion of high-volume observations into active governance workflows.SSRN