The Fly That Stopped: Mushroom-Body-Inspired Habituation as a Reward-Free Scheduling Prior for Autonomous Penetration Testing
A reward-free, fly-inspired scheduler reduced duplicate actions in lab autonomous penetration tests without finding more vulnerabilities.
Researchers evaluate a reward-free scheduler for autonomous penetration testing, inspired by Drosophila mushroom-body habituation, combining sparse state encoding with decaying counters over URL classes and tool families. In a second confirmatory stage, 8 of 10 screened lab targets remained measurable, and the scheduler lowered duplicate-action ratios in all 6 non-tied pairs, with an exact one-sided p-value of 0.015625. The largest reduction was from 51 to 18 duplicate steps inside a 60-step budget. The authors say this supports the complete scheduler on that population, not an isolated ablation or a gain in vulnerability discovery.
- A reward-free habituation scheduler penalizes repeated tool and URL-class selections.
- It reduced duplicate-action ratios in all six non-tied confirmatory target pairs.
- The largest drop was 51 to 18 duplicate steps within a 60-step budget.
- Authors report no vulnerability-discovery gain and no isolated habituation ablation.
- A 13,498-neuron circuit study found no action selectivity from local plasticity.
From the original article
1 min read at arxiv.orgAutonomous security-testing agents can spend much of a fixed action budget repeating earlier tool selections. We evaluate a reward-free scheduler inspired by mushroom-body novelty processing in Drosophila. It combines sparse state encoding with decaying habituation counters over structural URL classes and tool families.
By Theodoros Moutesidis · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.