DeepKeep has launched new guardrails for personally identifiable information that scan and filter personal data in prompts and responses across enterprise AI systems.
PII guardrails are installed within DeepKeep's existing AI firewall, which includes over 60 pre-built controls within the AI security platform. The company targets large organizations running generative AI and other machine learning tools across business functions.
DeepKeep says the new feature detects, edits and blocks personal data within AI workflows. It focuses on data that appears in prompts from employees and output generated from the model. According to the company, this is aimed at reducing the risk of unintentional data leaks as the use of AI in production environments expands.
Enterprise adoption of generative AI has raised concerns about how sensitive data enters and exits the models. Business users often paste inside information into chat-style interfaces, but models can surface or infer personal details from training data or integrated databases.
DeepKeep cited industry research showing that 69% of business leaders consider AI data privacy to be a top concern this year. The company said PII remains a prime target for cyberattacks and accidental disclosure, and many of the existing controls within organizations do not extend to model interactions and outputs.
regulatory pressure
PII guardrails are designed to be placed as a layer of security across existing AI operations. DeepKeep said it supports compliance with major data protection laws such as Europe's GDPR, California Consumer Privacy Act, Brazil's LGPD, and South Korea's PIPA. The product is focused on detecting sensitive data in context, not just in a fixed format, he said.
Regulators in multiple jurisdictions are increasing scrutiny of AI deployments that process personal data. Data protection authorities have issued guidance on training data, retention periods and individual rights requirements, and launched an investigation into how companies handle personal information within AI workflows.
DeepKeep says its systems evaluate PII in prompts and responses before reaching external models or internal applications. Guardrails then edits or blocks that information based on the configured policy. The company said this approach is aimed at reducing the risk of processing and reputational damage due to violations.
The company says the product covers categories such as names, contact details, government identifiers, financial information and other sensitive attributes. The company says customers can adapt detection and redaction policies to their own regulatory and business requirements.
Context-aware detection
DeepKeep compared the new guardrails to other tools on the market, such as Microsoft's Presidio. The company said its product detects a wider range of personal data categories and applies greater contextual accuracy when classifying.
The company says the guardrails use an AI model that identifies personal data even when it deviates from standard formats, such as common phone numbers or email addresses. The model said it applies understanding of context to determine whether data should be edited or left visible.
DeepKeep gave the example of an email address. This system blocks many instances of personal or internal email. If properly configured, it will allow public-facing enterprise support email addresses to pass through.
DeepKeep says internal testing and benchmark evaluations have shown it to have better data classification performance than competing tools. They report improved detection accuracy and fewer false alerts, which they say makes compliance workflows smoother for security and data protection teams.
The company says customers can deploy guardrails across multiple AI applications through a central platform. Security teams can view detection events and policy decisions in a single dashboard and adjust thresholds and rules as regulations and business needs change.
executive view
DeepKeep positions its platform as a way for enterprises to manage risk across the entire AI lifecycle, including large-scale language models and computer vision systems. Our products include AI firewalls, automated AI red teaming tools, and model scanning services.
Yossi Atlevet, chief technology officer at DeepKeep, said that as AI becomes more integrated into daily operations, there is a growing demand for stronger privacy controls.
“As reliance on AI continues to grow exponentially, enterprises are increasingly recognizing that they cannot afford to skimp on AI security,” said Yossi Atlevet, CTO of DeepKeep. “With our PII guardrails, businesses can take back control over their data security with faster and more granular layers of protection, ensuring privacy and compliance are built into every step of AI use. A continuous, intelligent layer of defense allows businesses to innovate responsibly and protect the data and trust of their employees and customers.”
PII Guardrails are immediately available within the DeepKeep platform and can be activated by existing customers from their dashboard.
