With their paper “Guardrail Vulnerabilities in Open-Source Language Models: Implications for Democratic Discourse and Marginalized Communities” Simon Münker and Fabio Sartori got a nomination for the best paper award at the 59th Hawaii International Conference on System Sciences (HICSS 59).
In their paper, the researchers investigate how widely used open-source large language models can still generate harmful or offensive content despite built-in guardrails. They show that, across seven prominent models, adversarial prompts repeatedly trigger biased or hateful outputs, posing risks to marginalized communities and democratic discourse. Their findings challenge developers’ safety claims. Especially Meta AI’s assertion that its models provide balanced perspectives on sensitive policy issues.
HICSS 59 is a leading multidisciplinary conference exploring information systems, software engineering, digital transformation, and socio-technical systems. It features a wide range of focused minitracks covering emerging technologies, communication, AI, sustainability, and organizational impacts. The conference is known for fostering dialogue across fields and connecting technical innovation with societal challenges.
