Share with your CISO
Enterprises deploying AI agents are hitting a data security wall that slows production rollouts more than model capability does. Protegrity, named Cloudera’s Emerging Technology Partner of the Year at EVOLVE26, argues that tokenization-based data protection is the unlock enterprises need to move agents from pilot to production at scale. Research commissioned by Protegrity from Enterprise Management Associates found 83% of AI project leaders report delays from governance processes, and 82% push agents live in a stripped-down state, missing the data access that would make them useful.
What this means for your business
The statistic worth sitting with is the 82% figure. If that number reflects your organization, your agents are technically in production but operationally neutered, running on sanitized inputs that strip out the customer, financial, or HR context that would make their outputs actionable. The security review process designed to protect the enterprise from data exposure is producing a different kind of risk: agents that can’t do the job they were funded to do, creating quiet ROI failure that rarely surfaces in a board deck.
Protegrity’s argument, delivered by a vendor with an obvious stake in selling data protection software, still holds structurally. Traditional encryption makes protected data mathematically unreadable, which means models can’t use it. Tokenization swaps sensitive values for stand-in substitutes that preserve format and usability while removing the underlying sensitive content. A Social Security number becomes a token that looks like a number, flows through model pipelines without triggering a compliance violation, and can be de-tokenized only by authorized systems. The agents get real signal; the sensitive value never leaves the protected perimeter.
The governance framing Diego Loureda offers is actually the sharper insight buried under the vendor pitch. Approving projects one at a time doesn’t scale when agents proliferate; approving patterns does. That’s a meaningful architectural shift in how security teams engage with AI programs. The CISO who rewires review processes around data handling patterns rather than individual use cases becomes an accelerant rather than a bottleneck. The one who doesn’t will watch business units route around security entirely, which is a worse outcome than any agent deployment risk they were trying to prevent.
Concept deep-dive: Tokenization
Tokenization replaces a sensitive value, say a credit card number or a patient ID, with a randomly generated substitute called a token that has no mathematical relationship to the original. Think of it as a valet key for data: the token works in the specific context it was issued for, but it can’t open anything else. Unlike encryption, there’s no algorithm to crack, which is why Protegrity claims quantum computing advances don’t threaten the approach. For AI pipelines, the business consequence is that models train and infer on tokens instead of raw sensitive data.
Based on reporting from Tokenization helps enterprises move AI agents into production, originally published 2026-11-01 09:25:00.
