The next stage of enterprise cybersecurity may depend on a capability that goes beyond identifying threats: understanding how an organization's technology environment works as a connected system.
A compromised identity, for example, can create risks that extend beyond the account itself. Determining which workloads it can access, how its permissions interact with infrastructure, and what could happen if those permissions are revoked requires an understanding of relationships across an enterprise environment.
Balestriero is an
At Sola, he will advise on research directions that connect these foundational AI concepts with the everyday challenges of enterprise security.
The Problem With Treating Security Findings in Isolation
Enterprise security environments contain interconnected identities, permissions, workloads, and infrastructure dependencies. Yet many security questions cannot be answered by examining a single suspicious event or configuration.
Identifying a compromised identity, for instance, does not automatically reveal the full extent of its access. Security teams need to understand which workloads that identity can reach and how its permissions relate to other components.
The response itself can introduce additional operational considerations. Revoking access may affect other systems or workloads, making it important to understand what could break before implementing a change.
This creates a distinction between detecting a security problem and understanding its implications.
Sola's approach centers on the idea that AI agents need a continuously updated representation of the environments they protect. Instead of processing disconnected findings, agents should be able to understand how different components relate and evaluate the potential consequences of proposed actions.
What World Models Could Bring to Cybersecurity
The concept of
Applied to enterprise security, these principles could help AI systems build comprehensive representations of organizational environments. An agent could potentially reason about how identities connect to workloads, what permissions actually enable, and how a security intervention might affect other parts of an organization.
This would change the role of an AI security agent.
Rather than simply responding to an alert, an agent with a deeper understanding of its environment could evaluate different possible actions and consider their downstream effects. The objective is to make decisions with greater awareness of operational context.
For example, an agent identifying a compromised identity would need to do more than flag the account. It would need to understand the identity's potential reach, assess the consequences of removing its access, and determine an appropriate response.
That distinction is important as enterprises increasingly introduce AI agents into security operations, where investigating threats, managing access, and carrying out remediation actions can involve interconnected systems.
Bringing Academic AI Research Into Enterprise Security
Balestriero's appointment brings expertise in representation learning and self-supervised learning into Sola's research efforts.
His work examines how AI systems develop structured representations of the world around them. Sola intends to explore how perspectives from this area of foundational AI research can inform the development of security agents capable of understanding complex enterprise environments.
Guy Flechter, Co-Founder and CEO of Sola Security, believes the industry needs to rethink what it expects from security systems.
"Cybersecurity has spent years building systems that can identify problems, but understanding the environment behind those problems is an entirely different challenge," said Flechter. “We believe the next generation of security agents will need something closer to a world model: an understanding of how systems connect, how changes propagate, and what the consequences of an action will be before it is taken.”
He also emphasized the value of bringing Balestriero's research perspective into the company's work.
"Randall brings a depth of expertise in how AI systems learn representations of the world. Bringing that perspective into cybersecurity is an important step for us. We want to build agents that don't just react to what they observe, but understand the environments they protect and can reason about the impact of their decisions."
A Research-Driven Approach to AI-Native Security
The appointment reflects Sola's position as a research-driven cybersecurity lab focused on the intersection of foundational AI research and enterprise security.
Rather than limiting innovation to adapting existing AI models for established security workflows, the company is exploring how advances in AI could enable systems to understand relationships, dependencies, and potential outcomes.
As AI agents take on more responsibilities in enterprise security, this distinction could become increasingly significant. Systems that understand the environments they operate in may be better positioned to assess risks and evaluate the consequences of their actions.
For Sola, bringing Balestriero onto its scientific advisory board is a step toward exploring that possibility.
This article was published as a press release by the author under HackerNoon's
