New agent-aware deterministic access controls designed to help enterprises prevent autonomous AI data exposure
Doha, Qatar, February 17th, 2026 – As enterprises accelerate the deployment of autonomous AI agents capable of independently accessing systems, sending communications, and moving data across cloud platforms, organisations are encountering new security challenges. Traditional access control designs break down, necessitating new agentic AI role-based access controls (Agenic RBACs).
At Web Summit Qatar today, Edison.watch announced the launch of its Deterministic agentic AI Security Framework, a security architecture designed to help enterprises prevent confidential data from being exposed by autonomous AI agents, even when those agents have legitimate system permissions.
The framework introduces agent-aware role-based access control (Agentic RBAC), enabling organisations to track data entering AI workflows and automatically prevent sensitive information from being transmitted outside approved boundaries.
Addressing a Growing Enterprise Security Gap
Autonomous AI agents can execute tasks across enterprise systems without direct human oversight. However, unlike human employees, AI systems lack contextual judgment and may combine or transmit data in ways that violate internal policy or regulatory requirements.
Enterprises report increasing incidents in which AI-generated communications inadvertently include sensitive internal information or aggregate data in unintended ways, due to broad system permissions granted to AI workflows.
“Enterprise security controls were built around human decision-making,” the Edison.Watch team said at Web Summit Qatar. “Autonomous AI follows instructions and uses available data efficiently, but without contextual awareness. That creates new categories of operational risk.”
Framework Capabilities
The Agentic AI Security Framework is designed to monitor and control data flow throughout AI-driven workflows rather than filtering outputs after generation.
Key capabilities include:
Data Intake Tracking
When an AI agent accesses information from sources such as databases, file storage, messaging platforms, or APIs, the system records the data source and classification.
Outbound Data Enforcement
If restricted data enters an AI workflow, the framework prevents unauthorised export through emails, messaging tools, file uploads, or external integrations while still permitting approved internal processing.
Real-Time Monitoring and Auditing
All agent actions are routed through a centralised gateway, enabling comprehensive audit trails and rapid identification of policy violations or abnormal behaviour.
This approach aims to allow enterprises to prevent data leaks proactively rather than identifying them after incidents occur.
Improving Visibility Into AI Activity
Organisations deploying AI agents often face fragmented visibility across platforms, including AI providers, collaboration tools, storage systems, and development environments.
Edison.Watch’s framework centralises activity monitoring, giving security teams a unified view of autonomous AI operations across enterprise infrastructure. The system can also help identify compromised credentials, unusual access patterns, or prompt injection attempts.
Security Considerations Slow AI Autonomy Adoption
While technology vendors have the technical capability to enable more autonomous AI agents, many enterprise deployments remain cautious due to concerns about data governance and regulatory compliance.
Despite this, organisations continue building autonomous workflows using APIs and integration tools to remain competitive and improve productivity.
Edison.Watch positions its framework as enabling enterprises to deploy AI automation while maintaining oversight and policy enforcement.
Deployment Options
Organisations implementing the framework can adopt one of two approaches:
Restricted Tool Usage
Limit AI agents’ ability to transmit data externally while still allowing internal data analysis.
Centralised Security Gateway Deployment
Implement a gateway layer governing all agent actions across enterprise systems like edison.watch on-prem or as SaaS, enabling full automation while enforcing agent-aware access controls.
Edison.Watch recommends gateway deployment for enterprises planning broader autonomous AI adoption and unlocking productivity gains, which integrate data with the existing pane of glass. (I.e. a SIEM)
Availability
The Agentic AI Security Framework is available immediately as part of OpenEdison, a freemium open-source AI data integration gateway platform. Documentation, deployment guidance, and policy templates are available through Edison.Watch.
For enterprise features like SSO, SIEM integration and others, enterprises can request technical demonstrations directly from the company.
About Edison.Watch
Edison.Watch develops deterministic security and governance solutions designed for autonomous AI operating in enterprise environments. The company focuses on enabling safe adoption of agentic AI by helping organisations control how AI systems access and transfer data across platforms.
