DeepKeep Joins the Microsoft for Startups Pegasus Program to Advance Enterprise AI Security

DeepKeep has joined the Microsoft for Startups Pegasus Program.

For DeepKeep, the timing is practical. AI applications and agents are moving into production while expanding beyond text into vision and other multimodal inputs across multiple languages.

These systems behave non-deterministically, process untrusted content, access sensitive data, and interact with external tools. They also change faster than conventional security review cycles can comfortably support.

Enterprise AI Security Extends Beyond the Model

Model-level safeguards are only one part of the security architecture.

The application or agent harness introduces retrieval pipelines, tools, third-party services, model artifacts, runtime policies, and access to enterprise data. This is where model behavior becomes an operational security problem.

A model might produce a safe response during evaluation and follow a different path in production. An agent can receive a legitimate request, retrieve a poisoned document or image, invoke an approved tool, and still create an unauthorized effect.

Controls can also behave differently across modalities and languages. A policy that detects a malicious instruction in English text may miss the same intent in another language, an image, or a video frame.

DeepKeep covers the AI security lifecycle from usage control and supply chain assurance to adversarial testing, agent risk assessment, and runtime protection across multilingual and multimodal environments.

The objective is straightforward: identify AI-specific risk before deployment, enforce policy while systems operate, and preserve useful evidence for security operations, incident response, governance, and compliance.

Why the Program Matters

The pressure is coming from concrete production requirements. Security teams must contain unsafe agent actions, protect regulated data across modalities and languages, and produce evidence for incidents and audits without breaking latency budgets.

The Microsoft for Startups Pegasus Program connects enterprise-ready startups with enterprise customers, Microsoft sales channels, go-to-market support, and technical guidance from Microsoft cloud architects.

Participation in the program will give DeepKeep more opportunities to work directly with enterprise security teams and production architectures.

That feedback validates the platform against real production conditions.

What Customers Should Expect

Customers should see benefits in three areas:

Clearer Production Deployment - Tested patterns for integrating DeepKeep into enterprise environments, including Azure-based architectures and other Microsoft cloud environments.

Broader Workload Validation - More validation across real AI applications and agents, including multilingual and multimodal workloads.

Operational and Compliance Evidence - Security findings, policy decisions, and runtime evidence that fit existing monitoring, incident response, governance, audit, and compliance processes.

DeepKeep's product direction remains unchanged: model-agnostic AI security, adversarial testing, contextual runtime enforcement, and practical evidence that security and compliance teams can use.

Built for the Part After the Demo

A convincing prototype is no longer enough. Enterprises need to understand what an AI system can access, how it can fail, which actions it may take, what will be blocked, and how an incident will be investigated.

The Microsoft for Startups Pegasus Program gives DeepKeep another way to test its platform against real production requirements and bring enterprise feedback directly into product development.