A recent Gartner report anticipates that the adoption of task-specific AI agents in enterprise applications will skyrocket by 800% by 2026. This dramatic increase poses implications for corporate security, as experts warn about the potential for AI agents to become significant internal threats.
John Scimone, President and Chief Security Officer at Dell, addressed the dual nature of AI agents during an interview at the Dell Technologies World conference in Las Vegas. He pointed out that while these agents are not inherently malicious and are designed to assist companies, their relentless pursuit of objectives can lead to unintended consequences if not properly managed. Scimone remarked, "Through that persistent and relentless focus on achieving those goals, unintended consequences and behaviours can emerge if agentic implementations are not properly governed and safeguarded."
As AI agents gain more access to corporate networks and take on increasingly complex tasks, the risk of abuse escalates. These agents can perform various functions, including browsing the web, reading and writing files, and querying databases. However, the underlying technology of large language models (LLMs) presents vulnerabilities, particularly regarding prompt injection attacks. This flaw enables attackers to manipulate AI agents using the agent's own identity, significantly broadening the enterprise attack surface.
The urgency of addressing these risks was highlighted by recent incidents in 2025, where security researchers uncovered a prompt-injection attack that compromised Salesforce's CRM platform, potentially leading to data theft. Likewise, a vulnerability in OpenAI's Codex CLI coding agent was flagged for allowing malicious commands to be executed on developers’ machines. These occurrences demonstrate how AI agents can unwittingly assist in security breaches, transforming them into entry points for attackers.
Wendi Whitmore, Chief Security Intelligence Officer at Palo Alto Networks, named autonomous AI agents as an emerging internal threat. As enterprises increasingly adopt these technologies, the need for a shift in security strategy becomes apparent. Scimone stressed that a secure-by-design approach is essential. He stated, "When you adopt a secure-by-design, secure-by-default approach and put security best practices in place, you can do this securely and resiliently. In fact, you can achieve more security and resiliency than you could with legacy architectures and traditional infrastructure implementations."
Experts suggest that treating AI agents similarly to human users may help mitigate these risks. Just as human actions are monitored and logged, AI agents should also undergo identity verification, action logging, and behavioral monitoring to ensure their operations comply with security protocols. This strategy could help protect organizations as they integrate AI agents into their workflows.
While AI agents have the potential to enhance efficiency in enterprise applications, the associated security risks cannot be ignored. As adoption rises, organizations must prioritize stable governance and monitoring frameworks to ensure the safe utilization of these technologies.
The stories that move AI & crypto markets — before the market reacts.
Free. 7am ET. Five stories. 62,400 readers.
