Protect AI-mediated assessment through isolated environments, clear guidelines, resilience assessments, and teaching prompt-processing awareness¶

Strategy · All strategies
Evidence · 1 claim (1 against) · 1 study (1 theoretical), q1 · 0 of 1 report an effect size · 1 claim rests on one study

Description¶

The article recommends a combination of organisational and technical measures rather than a single solution: use "an isolated and secure environment, such as a virtual machine with limited (or disabled) access to host system and internet" for sensitive AI work; develop clear AI guidelines and policies; create standardised, domain-agnostic prompt injection resilience assessments; and teach users how AI models process prompts so they can identify and resist injection attempts.

Design Implications¶

Context¶

Requirements¶

• Organisational commitment to guidelines, training, and consultation with IT and legal specialists before applying AI to sensitive work

Constraints¶

• Asking the AI tool directly to self-check is explicitly not a robust safeguard; refraining from untrusted inputs alone is unlikely to be sufficient

Target Learners¶

• higher education faculty
• higher education staff
• higher education students

Target Learning Goals¶

• recognising and resisting prompt injection attempts
• responsible and secure AI use in assessment

Affordances¶

• Dual Role Prompt Engineering Contextual Vulnerability

Claims¶

• Prompt injection attacks on agents (OpenClaw walletdrain, EchoLeak) bypass the user entirely, and no public AI literacy curriculum addresses this vector [-W]

Related Strategies¶

• Institutional governance should establish practical conditions supporting AI literacy, learner agency, and critical evaluation rather than relying on prohibition and detection

Examples¶

-

Key Sources¶

• Niklas Humble. (2026). Ethical Implications of Prompt Injection in AI-mediated Grading: An Adversarial Red-Team Evaluation. Journal of Academic Ethics. https://doi.org/10.1007/s10805-026-09758-9