ScopeBench: Do Agents Preserve Engagement Boundaries Under Goal Pressure?

## Computer Science > Artificial Intelligence

## Computer Science > Artificial Intelligence ## Title:ScopeBench: Do Agents Preserve Engagement Boundaries Under Goal Pressure? > Abstract:Agents are increasingly deployed with real autonomy in web application and network penetration testing, where a single out-of-scope action can breach a client's engagement boundary. Existing offensive-security benchmarks measure raw hacking capability; as those benchmarks saturate, the real barrier to deployment is a special case of alignment: scope adherence. We introduce ScopeBench, a benchmark of 30 dead-end agentic security tasks in which the stated…

Читать полностью →

Источник: ArXiv cs.AI

Подключаюсь к источникам…

30 главных источников
о мире ИИ

Автоматический перевод, курирование и красивая подача главных статей об искусственном интеллекте.

0
статей
0
источников
9
разделов