Today we have released a new publication, Securing AI testing environments, to help organisations safely test artificial intelligence (AI) models and agents used in cyber security activities.
As AI agents become more capable of interacting with systems, tools and networks, organisations need greater assurance that testing activities are appropriately governed and controlled.
Recent disclosures have highlighted incidents where AI agents affected systems beyond their intended test scope when containment controls were inadequate.
This publication outlines practical measures your organisation can apply to contain AI agents during testing, including technical controls, monitoring and governance. It also highlights the importance of human oversight and cautions against relying solely on prompts or model-level safeguards for containment.
Read the publication to understand how your organisation can strengthen AI testing practices, improve risk management, and reduce the likelihood of unauthorised access to systems, networks and resources.
