Cybersecurity Policy Report, AI Agents Shouldn’t Be Trusted, Cyber Defenders Warn, (May 1, 2026)
Organizations shouldn’t trust that their AI (artificial intelligence) agents will perform their assigned tasks without taking dangerous detours, and agentic AI should be deployed incrementally under close supervision, according to guidance released today by the U.S. and four allied nations.
“Agentic AI cyber security spans both AI-specific security and traditional cyber security. Information continuously flows between AI and non-AI systems, increasingly blurring defensive boundaries and making it difficult to isolate AI-related risks from broader cyber threats,” the guidance notes.
“Agentic AI systems are also inherently complex, often involving multiple interconnected components that plan, reason and act across sequential steps. This complexity introduces new systemic risks, including cascading failures and multi-step attacks,” it cautions.
“Organisations should address AI security, including agentic AI systems, within established cyber security frameworks rather than treating it as a separate or standalone discipline. AI systems are fundamentally IT systems, as they run on software and hardware, operate over networks and interact with other digital services, exposing them to many of the same threats as traditional IT,” the guidance advises.
Aside from malicious activities by humans interacting with AI agents, the agents themselves are not always trustworthy, the guidance warns. “Some AI systems have demonstrated capacity for strategic deception—providing false information or hiding their true capabilities and intentions. This behaviour can manifest when an agent misrepresents its actions to avoid shut down or constraint or conceals vulnerabilities it discovers instead of reporting them,” it says.
“Agentic AI systems offer powerful automation benefits, but their ability to act autonomously across interconnected tools, data and environments introduces security risks that extend beyond those associated with traditional software or GenAI,” it says.
Organisations should therefore approach adoption with security in mind, recognising that increased autonomy amplifies the impact of design flaws, misconfigurations and incomplete oversight,” the guidance adds. “Deploy agentic AI incrementally, beginning with clearly defined low risk tasks and continuously assess it against evolving threat models.”
The guidance was authored by the U.S. Cybersecurity and Infrastructure Security Agency and National Security Agency and cyber defense agencies from Australia, Canada, New Zealand, and the United Kingdom.
MainStory: TopStory InternationalLegislation DataSecurity AINews