Dymium GhostAI gateway aims to govern enterprise AI access

Dymium launched the Dymium GhostAI gateway today, a security and governance product for enterprise AI use, with pricing based on consumption and no specific rates disclosed. The company says GhostAI gives security teams a control point between company data and the AI models, agents and tools trying to use it, a problem that grows as employees bring proprietary information into AI workflows.

GhostAI is available through an early-access program. Dymium said general availability is planned, but it did not provide a date. Founder and Chief Executive Denzil Wessels said customers have used the underlying technology for about six months and claimed most organizations can set it up in less than five minutes.

What does Dymium GhostAI do?

GhostAI sits between enterprise systems and AI services, inspecting requests and responses, enforcing policy and recording activity for audit purposes. Dymium positions it as a single policy layer for four areas it labels Models, Context, Tools and Data, and claims it is the first secure AI gateway to govern all four through one engine rather than a bundle of separate products.

The pitch addresses a practical tension for enterprise AI buyers. Foundation models become more useful when they can access internal data, but that access can expose private information, regulated records, credentials or intellectual property. Wessels said Dymium built GhostAI for that point of failure, where sensitive company data is introduced into model workflows.

How the model and data controls work

At the model layer, Dymium says GhostAI can route traffic across more than 800 public and private models. Users can pick a model themselves, or policies can determine where a prompt should go based on the requested task, the sensitivity of the data and governance rules. Dymium said confidential requests could be directed to private inference through Amazon Web Services Bedrock, Google Vertex AI or an enterprise data center.

For data protection, GhostAI can identify sensitive material before it reaches a model, including names, account numbers, credentials and proprietary content. Depending on the policy, the system can block, mask or redact the information. Dymium also says it can substitute synthetic values that preserve analytical relationships while keeping the original data inside the organization.

After a model responds, GhostAI can retrieve the real values from a protected vault and reinsert them for authorized users. Dymium also describes the system as using a zero-copy architecture, meaning AI tools can work with live enterprise data without first duplicating or staging it elsewhere.

Context, tools and early use cases

The context layer includes shared memory for moving conversations and organizational knowledge among users, models and agents. Dymium says policies can limit that memory by company, group, topic or individual user, while sensitive information is kept separately using token and vault protections.

The tools layer covers API calls and interactions using Model Context Protocol, the standard gaining use for connecting AI systems to external software and data. Dymium said GhostAI can log each tool call and data operation for auditing.

Initial supported use cases include secure chat, coding assistants and agentic workflows. Dymium is selling first to regulated sectors such as financial services and healthcare, along with companies trying to protect intellectual property. The company did not disclose customer names, revenue, headcount or funding details in connection with the launch.

This story draws on original reporting from SiliconANGLE.