The rapid integration of generative AI (e.g., ChatGPT) into STEAM education presents a critical challenge: harnessing its potential as a learning partner while mitigating risks to academic integrity. Traditional approaches that simply prohibit or block AI use are impractical and fail to teach responsible engagement. This paper introduces SteamGPT, a novel, ethics-aware platform that provides real-time, ethical guidance to transform generative AI from a shortcut into a structured learning aid. Built on a FIPA-compliant JADE multi-agent framework, SteamGPT coordinates a suite of intelligent agents: monitoring agents analyze linguistic and behavioral patterns, decision-making agents apply adaptive integrity policies, and coordinator agents ensure system-wide fairness and transparency. Crucially, this guidance is delivered on a strict privacy-by-design basis. The client-side interface operates with user anonymity, employing local on-device pseudonymization and secure, attribute-based encryption. This architecture allows the system to process interactions and deliver contextual feedback based on session tokens, not personal identifiers, ensuring student privacy is never compromised. A pilot study in a university STEAM course demonstrated the platform's dual efficacy. SteamGPT significantly reduced instances of AI-assisted plagiarism while simultaneously improving students' problem-solving and critical reasoning skills. This work proves that academic integrity can be fostered without surveillance, offering a scalable, ethical framework for deploying trustworthy AI services in education. SteamGPT stands as a model for safeguarding integrity while fostering a culture of responsible innovation.

AI in Education, Multi-Agent Systems (MAS), Academic Integrity, Privacy-preserving AI, STEAM Education, Ethical AI, JADE, Responsible AI Adoption