Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.30614.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Research paper defining the 'authorship hazard' in agentic dataspaces: LLM agents that are both policy subjects and policy authors can draft norms that bind themselves. Proposes a principle that agents are subjects of governance, never authors, and reports empirical results showing 80 authorization reversals on frozen agent drafts and 105/105 vs 0/105 execution-boundary enforcement depending on constraint compilation.

Names a concrete governance failure mode in agent systems, backs it with specific experimental numbers, and proposes an enforceable architectural principle before adoption widens.

Cited by

No citations on record.