Authorities in Abu Dhabi are implementing enhanced cybersecurity measures to protect over 50 government entities as the use of artificial intelligence (AI) systems expands across public services. Officials caution that AI agents—autonomous programs capable of performing complex tasks—could be manipulated by malicious actors to access sensitive information, including financial records and personal data, or even conduct unauthorized transactions.

Najyb Al Maskari, executive director of the Government Cybersecurity Operations Sector at the Department of Government Enablement (DGE), emphasized that the principal threat lies not in AI errors but in attackers directing AI systems to carry out harmful activities under the guise of authorized commands. He highlighted the importance of a comprehensive security framework that goes beyond a single layer of protection.

“We are securing all 50 entities,” Al Maskari said, underscoring the government’s strategy to create an AI-native infrastructure with “defense in depth.” This approach includes continuous oversight, auditing, real-time monitoring, strict guardrails, and the capacity to deactivate AI agents if they exhibit suspicious behavior. He warned of risks arising when AI models and their agent programs are deployed without sufficient supervision, noting that agents left to operate unchecked pose significant cybersecurity challenges.

Supporting these concerns, Dr. Najwa Aaraj, CEO of the Technology Innovation Institute, remarked at the AI Everything conference that AI agents executing tasks without real-time monitoring could effectively function as malware within systems. Aaraj advocated for stronger regulatory standards, such as clearly identifying AI systems and imposing strict controls on their activities and access rights.

This cautionary stance is echoed by cybersecurity experts worldwide. Vladislav Tushkanov, head of the Kaspersky AI Research Centre, explained that AI agents have become appealing targets for cybercriminals because they often possess machine-level access to corporate networks, credentials, and confidential data. He described malicious "skills" — specialized code modules enabling AI agents to perform particular functions — as a potential vector for embedding harmful instructions.

Tushkanov detailed a novel tactic known as “living off local agents,” where attackers leverage existing AI agents installed on machines to autonomously seek out valuable information, reducing the need for the attacker’s direct involvement. He also noted that AI's capabilities have lowered technical barriers for cybercriminals, facilitating an increase in the volume and speed of attacks.

Kaspersky’s research has uncovered instances of attackers impersonating popular AI platforms to spread malware and distributing counterfeit software packages aimed at AI developers—software that appears legitimate but contains hidden malicious components.

The evolving landscape underscores the urgency for robust, multilayered defenses and vigilant monitoring as government and corporate sectors accelerate AI adoption in their operational frameworks.