As AI agents become more sophisticated, there is growing interest in endowing them with internal state representations analogous to affective states. However, affective states without regulation can lead to instability, perseverative loops (rumination), and vulnerability to manipulation. We introduce the Affective Regulation Core (ARC), a control framework inspired by prefrontal cortex functions that maintains stability in agents with internal affective states, together with the Affective Stability & Safety Benchmark (ASSB), a reproducible evaluation protocol. Across 6 research lines and 15 controller architectures (P, PID, LQR, LQI, hierarchical, meta-control, H-infinity robust, and adaptive variants), controllers with integral action or H-infinity robust design drive the Rumination Index to zero while maintaining PerfMean ≥ 0.93. H-infinity robust controllers are the most consistent architecture across the full suite, including adversarial coupling, where integral controllers collapse due to integral windup — an important negative finding reported transparently. All code, data, and per-seed results are released at https://github.com/edamianreynoso/arc-assb-controller under Apache-2.0 license (tag arc-paper-v1). A companion paper (in preparation) deploys the same controllers inside a live LLM-based cognitive agent and empirically validates the predictions of this work under naturalistic adversarial conditions.
J. Eduardo Damián Reynoso (Sun,) studied this question.