Tag
This paper introduces Anchor, a synthetic audit framework to evaluate long-horizon persona stability and trajectory recall in AI companions, finding that current models fail to reliably preserve identity and memory across extended interactions.
A practitioner shares challenges and tools for monitoring autonomous AI agents in production, covering runtime prompt injection detection, tool-call auditing with reasoning traces, behavioral drift detection, and multi-agent authorization, while testing tools like Arize Phoenix, Protect AI Guardian, Metoro, Alice, Asqav, and Microsoft Agent Governance Toolkit.
MemEvoBench introduces the first benchmark for evaluating memory safety in LLM agents, measuring behavioral degradation from adversarial memory injection, noisy outputs, and biased feedback across QA and workflow tasks. The work reveals that memory evolution significantly contributes to safety failures and that static defenses are insufficient.