Protegrity launches unified policy engine for enterprise AI safety
The new tool helps companies give AI agents access to internal data without exposing sensitive personally identifiable information.
The short version
- The new unified policy engine controls access for people, AI agents, and applications using the open-source Cedar policy language.
- Protegrity's PII detection model identifies 69 types of sensitive data, including health conditions and cryptocurrency wallet addresses.
- Research shows 83% of organizations have delayed AI deployments due to security or compliance reviews.
- The new AI Gateway acts as a centralized orchestration point to check prompts and responses across AI workflows.
- The solution currently runs on Microsoft Azure and Amazon Web Services, with Google Cloud and on-premises support coming soon.
Why it matters
Most companies struggle to move AI from testing to production because they fear agents will leak private data or perform unauthorized actions. This tool provides a way to give AI the context it needs to be useful while maintaining strict security boundaries.
Bridging the AI security gap
Many organizations are currently stuck in a difficult position when trying to adopt artificial intelligence. While AI is more valuable when it can access internal business context, that same context often contains a company's most sensitive information. Security teams are often forced to choose between making an AI tool useful or keeping it safe from data leaks.
According to research by EMA’s State of AI Friction, these concerns have real-world consequences for business growth. The study found that 83% of organizations have delayed their AI production deployments specifically because of security or compliance reviews. Furthermore, 82% of companies have had to deploy AI with reduced functionality because they could not resolve the underlying risks associated with the technology.
Protegrity is attempting to solve this friction by creating what they call a secure context layer. The goal is to ensure that every AI agent receives exactly the information it needs to complete a task while being blocked from doing anything it is not authorized to do. This approach aims to move AI out of the experimentation phase and into full production environments.
How the policy engine works
The core of the new update is a unified policy engine built on the open-source Cedar policy language. This engine allows security teams to apply specific rules to users, AI agents, and the queries they submit. It works by integrating with a model that detects personally identifiable information (PII), which is any data that can identify a specific individual.
When an AI system processes a request, the engine determines what the specific user or agent is allowed to see and which actions they can perform. This is particularly important because many modern agents are now capable of writing to databases or triggering external application programming interfaces (APIs). Protegrity's architecture applies these controls across the sensitive data supplied to the system, the permissions of the actors, and the actual prompts and responses generated.
To improve accuracy, Protegrity included a fine-tuned PII detection model that recognizes 69 different types of sensitive information. These categories include specific items like biometric data, health conditions, and cryptocurrency wallet addresses. The company reported that this model achieved a precision of 0.902 and a recall of 0.910, which they claim are higher than results from several major providers including Amazon, Microsoft, and OpenAI.
New tools for security teams
To help security teams manage these complex rules, Protegrity introduced a new interface called Data & AI Fast Lanes. This tool allows teams to detect and classify sensitive information before it ever reaches an AI model. By identifying this data early, companies can create protection policies and apply controls at the source.
The release also includes a new component called AI Gateway. This serves as a centralized orchestration point that monitors prompts and responses throughout various AI workflows. It allows organizations to run Protegrity’s specific guardrails alongside the security controls they already have in place. This ensures a consistent layer of protection regardless of which specific AI tools are being used.
Additionally, the update includes two features currently in technology preview: Secure Query and Iceberg Protector. Secure Query allows for searching and filtering protected data without needing complex database triggers. Iceberg Protector extends these protections to Apache Iceberg, which is a common open-source framework for managing large datasets. These tools are already being used by customers to provide feedback before they reach general availability.
Deployment and future roadmap
The new capabilities are currently available in the Protegrity AI Team Edition. These features run on a Kubernetes-based foundation known as the Protegrity Provisioned Cluster. This allows customers to deploy and operate their security tools within a single cloud environment on either Microsoft Azure or Amazon Web Services.
Protegrity has announced plans to expand the reach of these tools in the coming months. They intend to add support for Google Cloud and on-premises environments, such as Red Hat OpenShift. This expansion is designed to give a wider variety of enterprises the flexibility to choose where their secure AI infrastructure lives.
This release represents only the first phase of a larger roadmap for the company. Future updates will extend these secure context controls to include the specific intent of AI agents. They also plan to monitor and secure the communication that happens between different agents as they work together to complete complex tasks.
What to watch
- The release of Secure Query and Iceberg Protector from technology preview to general availability.
- The expansion of the platform to support Google Cloud and on-premises environments.
- How the platform handles 'agent intent' in future updates.
Try it
Check if your organization's AI deployment has been delayed by security reviews; if so, look into whether your current controls are designed for humans or for AI agents.
Sources and links
- OfficialEMA’s State of AI Friction protegrity.com
- Officialprotegrity.com/news/protegrity-brings-secure-context-to-enterprise-ai- protegrity.com
- ReportingCybersecurity Insiders: the article we sourced
First reported by Cybersecurity Insiders. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.