ERTS: Adversarial Robustness Testing of Ethical AI via Semantic Perturbation in a Bounded Consequence Space

A new adversarial testing framework called ERTS evaluates whether AI systems can maintain ethical reasoning under deliberate manipulation, according to a paper posted to the arXiv preprint repository. Only one-third of the models tested passed the assessment [1]. The Ethical Robustness Testing System, introduced in a paper submitted June 11, 2026, encodes ethical dilemmas into a 22-dimensional Ethical Consequence Space grounded in established ethical theory [1][2]. It then applies 17 semantic perturbation functions, which are governed by 6 validity constraint classes, including a novel semantic coherence constraint [1][2]. The framework measures how far a model’s decisions deviate under attack using a 4-component Ethical Instability Index [1][2]. Researchers evaluated four structured baseline models and two production large language models — Gemini 2.0 Flash and Llama 3.2 — across 50 ethical scenarios spanning eight deployment domains, generating 1,500 adversarial test cases [1][2]. Large language models are neural networks trained on vast text corpora and serve as the foundation for modern chatbots, though biased or inaccurate training data can undermine their reliability [8]. Only 33 percent of the models achieved assessment clearance [1][2]. The local Llama-3.2 model showed particular vulnerability to fairness corruption and information degradation attacks, recording an evaluation result of 0.737 [1][2]. The paper states that no existing framework combines a bounded ethical consequence space, semantic coherence constraints, and domain-adaptive assessment in a single adversarial testing pipeline [1][2]. The paper appears on arXiv, an open-access repository of electronic preprints that has been operating since August 1991 and now receives roughly 24,000 submissions per month [6]. Papers on arXiv are moderated but not peer-reviewed before posting [6]. The repository hosts work across mathematics, physics, computer science, and related fields [6].