Benchmark RadarAI BENCHMARK PROFILE
MemSecBench
Evaluates lifecycle security of agent memory systems with 310 cases across 48 contexts, using a Write-Execute-Forget protocol and evidence-based adjudication across seven checkpoints.
- Released
- 2026-07-29
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Provides insight into how malicious instructions can persist in memory systems and affect later actions, highlighting security differences across configurations.
Motivation
Memory systems allow agents to retain and reuse information from past interactions, but they can also let malicious content persist.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.