ZeroDrift launches Guard for Agents compliance service

ZeroDrift's launch extends its compliance technology, previously centered on Command, into the agent workflow space. The company positions Guard as sitting between the generation of a message and its delivery, catching violations in real time rather than after the fact. This pre-send positioning is notable because it targets a specific moment in the AI content pipeline where policy gaps commonly occur.

The business model reflects an operating-cost approach: one cent per validation, with checks covering email, text and Slack posts. Implementation is designed to avoid requiring developers to rewrite policies in proprietary languages — instead, a small language model identifies rules from imported policy documents, and customers can inspect and select rules in a visual interface. The underlying model, Anchor 3.0-Preview, was post-trained on attorney-labeled data, and the company logs enforcement decisions for audit purposes.

Aroomoogan indicated the company is considering expansion beyond written communications into voice agents, responding to demand in financial services and healthcare. The technology also works across conventional marketing and CRM systems without a language model, though rewrites are limited to machine-generated content. Notable for potential enterprise buyers, ZeroDrift has not yet released latency measurements or independent performance results, though it plans third-party benchmarks in the future.