Project-by-project security reviews can’t keep pace with enterprise AI agents

Enterprises racing to deploy AI agents are discovering that data protection measures like tokenization significantly influence project timelines. However, it is challenging to retrieve sensitive data once it has been fed into models. Security reviews designed for a slower time are impeding these deployments. The focus is shifting from data protection to ensuring usability of the protected data.

Diego Loureda, vice president of customer success and presales at Protegrity USA, highlighted that data leaks remain a concern when sensitive data is incorporated into models, vector stores, or agent memory. He emphasized that removing such data is nearly impossible. Traditional encryption methods can make data too restrictive for analysts and models to use.

Protegrity's approach leverages tokenization, which replaces sensitive values with substitutes. This method is also ahead in preparing for quantum computing threats. Research commissioned by Protegrity found that 83% of leaders reported delays in projects, and 82% of leaders deploy agents into production with limited capabilities.

Tokenization allows security teams to facilitate the deployment of agents with access to necessary tools and data, enabling them to derive value from AI.