AI governance starts with permissioned data

AI governance is the set of policies, controls, and evidence that determine how AI systems use data, plus the machinery that enforces them. In practice it has to do four things: capture where AI touches enterprise data, govern the purposes and permissions behind each approved use case, enforce those rules at runtime, and audit every decision. Frameworks and assessments cover the paperwork; Ketch AI Governance covers all four.

They're the two products behind Ketch AI Governance, and they split the job. AI Sentry is the enforcement engine: it sits inside your AI applications, inspecting prompts, detecting personal data, and blocking or redacting PII in real time. The Ketch Agent Network automates the privacy program around it, drafting DPIAs, ROPA records, and policy updates for your team to review and approve. AI Sentry controls what happens at inference time; the Agent Network keeps the program current. Both run on the same Permission Vault, Identity Sync, and Transponder.

AI Sentry inspects prompts, responses, and outbound transfers for personal data that isn't permitted to move: names and identifiers, financial details, health attributes, and biometrics. Typical catches include PII pasted into a support copilot's prompt, sensitive fields headed to an ad partner, and agents reaching for records outside their approved purpose. Depending on your policy, AI Sentry redacts the field in flight, blocks the request, or scopes the data down.

AI Sentry sits between your applications and your models, so it works with the AI stack you already run rather than replacing it. Enforcement happens at the point of use, which means no model retraining and no per-application configuration: define a policy once in Ketch and it applies across every model, agent, and copilot that touches your data.

The opposite, in practice. Teams stall on AI when every use case needs a manual privacy review, and they move when permission questions answer themselves. With consent and policy enforced automatically at the data layer, approved use cases ship, non-permissioned data is blocked by default, and the evidence record builds itself instead of waiting on a quarterly audit.

AI Sentry inspects every prompt and model call for personal data before it leaves your environment. When sensitive fields appear in a request the calling agent isn't permitted to see, AI Sentry redacts them in flight or blocks the request entirely. Rules like "no health data without explicit consent" or "SSN never accessible to sales agents" are defined once and enforced consistently across every model, agent, and application, with no per-system configuration.

Dirty data is what accumulates when information flows into applications and AI models without consent context, purpose limitation, or accountability. When it surfaces in a regulatory inquiry, a customer complaint, or a discovery request, the cost isn't just legal: models get retrained from scratch, datasets get rebuilt, and quarters get lost. Ketch addresses it at the source. AI Sentry checks consent and purpose limitations at every inference call, so the data reaching AI is permissioned by default rather than cleaned up after a violation.

Every decision AI Sentry makes, from blocked prompts to redacted fields to allowed queries, is logged with full audit context: who made the request, what data was involved, when it happened, and which policy applied. The logs are designed to answer the questions regulators are starting to ask: can you prove how data is used, and can you control it in real time?