Benchmark RadarAuthMem-Bench
AuthMem-Bench evaluates authority collapse in persistent memory for LLM agents. It uses a paired benchmark holding claims and tasks fixed while varying source authority, measuring write-time collapse, authorization errors, and automatic authority preservation.
- Released
- 2026-08-03
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Memory consolidation can erase authority constraints, leading to unauthorized actions. AuthMem-Bench provides a controlled benchmark to measure and improve authority preservation in memory systems, relevant for safe agent deployment.
Motivation
Persistent memory allows (self-evolving) LLM agents to adapt across tasks by consolidating heterogeneous interaction histories into reusable facts, preferences, observations, and rules.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.