AI agents have been a key feature in recent innovative product launches, but cybersecurity experts hold varying opinions on these technologies.

A study by SailPoint revealed that 96% of tech professionals consider AI agents as an increasing security threat. However, nearly all respondents planned to expand the use of agentic AI in the upcoming year.

The primary concerns raised by respondents included the agents’ access to protected data, the risk of unintended actions, and the possibility that an AI agent could share sensitive information without permission.

Issues Surrounding Data and Privacy

In generative AI platforms, models often produce inaccurate or false conclusions due to the ongoing “black box” issue, making it difficult for analysts to understand why an AI made a wrong decision.

Privacy remains a significant concern in AI systems that require extensive data. Even though well-established gen AI platforms like ChatGPT are built with mechanisms to protect sensitive data, AI agents frequently need access to private information such as financial details to perform their tasks.

A troubling finding from the SailPoint study indicated that just under a quarter of respondents reported instances where AI agents had been manipulated into revealing access credentials.

Additionally, 80% of respondents discovered that their companies’ AI agents were performing unintended actions, such as unauthorized system access, releasing protected data, and retrieving inappropriate content.

The Dawn of Agentic Commerce

In the face of these concerns, agentic commerce is still advancing. Visa and Mastercard have launched platforms aimed at transforming AI agents into personal shoppers capable of making purchases with minimal user interaction.

PayPal quickly followed suit by integrating its payments directly within Perplexity’s chat interface as part of their latest partnership in agentic commerce.

In light of this potential, more initiatives across various industries, including cybersecurity, are expected to emerge. However, organizations must continuously prioritize privacy and security measures for these AI agents.

The SailPoint study highlighted that 92% of respondents believed governing AI agents was crucial for enterprise security.