MCLEAN, VA — September 23, 2026 — Replica Cyber, creator of the patented Secure Operations Platform for high-risk work, today announced the successful completion of an advanced cybersecurity assessment conducted through the U.S. Defense Innovation Unit (DIU).
The assessment wasn't designed simply to scan Replica for vulnerabilities. Testers targeted internal infrastructure, attempted privilege escalation and account exploitation, and tested whether activity inside Replica environments could move laterally or compromise the infrastructure underneath them.
They could not break out.
The implications are ground breaking. As organizations adopt agents at an accelerated rate existing guardrails are insufficient to ensure tracebility and containment. With Replica’s proven secure environments we remove the threat of rouge agents breaking out of intended scope.
Even after testers achieved root-level privileges inside a guest environment, testing verified that those privileges did not enable escape from the environment or compromise of the underlying host infrastructure. Attempts to use credentials obtained from inside an environment to reach protected cluster, code and data resources were also blocked by Replica's security controls.
For enterprises racing to adopt AI, that result points to a fundamentally different security model:
Replica lets you operate differently.
Instead of requiring organizations to trust every AI model, agent, application and dataset before allowing it to operate, Replica gives organizations a place to run technology they may not trust—without extending that trust to the enterprise around it.
That means organizations don't have to choose between moving at AI speed and maintaining enterprise security.
They can do both.
"AI changes the security assumption," said Kristopher Schroeder, Co-Founder and CEO of Replica Cyber. "With the proliferation of agents You can't assume every model, agent, application or piece of code will always behave exactly as intended. Replica lets you operate differently. We give AI the freedom to work while putting hard boundaries around where that work can go. The Defense testing demonstrated exactly why that architecture matters—even when an attacker achieved elevated privileges inside an environment, they couldn't turn that access into compromise of the platform around it."
Assume the AI Could Be Compromised. Design So It Doesn't Matter.
The traditional enterprise security model depends heavily on establishing trust before technology is allowed to operate: approve the application, approve the infrastructure, approve the model, approve the data, approve the connectivity.
AI makes that increasingly difficult.
Models change. Agents execute actions autonomously. Open-source technologies can introduce unknown dependencies. AI applications increasingly combine sensitive enterprise data with external models, APIs and rapidly evolving software.
Replica takes a different approach.
Don't make perfect trust a prerequisite for innovation. Control the environment in which innovation happens.
Replica creates isolated, governed environments where AI models, agents, applications, data and users can operate together—while security controls determine what can enter, what can leave, what resources can be reached and how the environment interacts with the enterprise.
The Defense assessment provided a powerful demonstration of that architecture.
The assessment found that Replica's network architecture partitions infrastructure into distinct isolated networks, while IAM controls blocked attempts to use guest credentials to reach unauthorized resources.
And when testers achieved root privileges within a guest environment, that compromise remained contained. The assessment specifically found that root access did not provide a path to escape the environment or compromise the underlying host infrastructure.
Compromise inside the environment did not become compromise of the platform.
That is the security assumption Replica brings to AI.
Already Running AI Inside Some of the World's Largest Banks
This isn't a future architecture.
Four of the world's ten Global Systemically Important Banks (G-SIBs) already use Replica to isolate and execute proofs of concept, emerging technologies and AI initiatives without waiting for traditional infrastructure deployment or creating security exceptions.
Replica gives those organizations an immediately available operating environment with governance and controls built in from the beginning.
Teams can bring in new models. Run agents. Work with sensitive data. Test emerging technologies. Execute proofs of concept. Move from experimentation toward production.
And they can do it without turning every new technology into another exception for security to manage.
The controls behind that capability weren't originally designed for low-risk experimentation. They were developed to support some of the most sensitive digital national security work—where isolation, attribution, governance and control aren't optional.
Today, that same architecture is being applied to enterprise AI.
From "Is This AI Safe?" to "Where Can This AI Operate Safely?"
For security leaders, Replica changes the question.
Instead of asking:
Can we trust this model? Can we trust this agent? Can we trust this open-source technology? Can we trust every action it might take?
Organizations can ask:
Where can we allow it to operate safely?
Replica provides that place.
The platform's architecture is supported by 24 global patents and combines environment isolation, policy enforcement, identity, network controls, data governance and environment automation into a unified Secure Operations Platform.
For organizations adopting AI, the result is a simple proposition:
Assume it could be compromised. Assume it could behave unexpectedly. Let it work anyway.
Because when the environment—not blind trust—is the security boundary, organizations can move faster without accepting uncontrolled risk.
"The companies that win with AI aren't going to be the ones that figure out how to eliminate every risk before they move," Schroeder said. "They're going to be the ones that build an operating model where they can move despite the risk. That's what Replica enables."
Empower the Work. Eliminate the Exceptions.
Replica was built around a simple principle: security should enable critical work, not become the reason it can't happen.
As AI dramatically increases both the speed of innovation and the potential consequences of compromise, that principle matters more than ever.
Replica lets organizations operate differently: isolate the risk, govern the environment, and let people and AI move at the speed the mission demands.
About Replica Cyber
Replica Cyber created the patented Secure Operations Platform for high-risk work. Originally engineered to enable highly sensitive national security operations, Replica provides organizations with governed environments where people, applications, data, AI and emerging technologies can operate without exposing the broader enterprise to unnecessary risk.
Supported by 24 global patents and deployed across government and some of the world's largest financial institutions, Replica enables organizations to investigate, innovate and operate without creating the security exceptions that slow the business down.
Empower the Work. Eliminate the Exceptions.