01Prevent
Agent control platform
Identity, permissions, policy management, and MCP enforcement control which capabilities an agent may reach.
Platform →Averon/Company
Build the control. Test the boundary.
Averon builds the identity, policy, and enforcement layer for AI agents. The company also performs adversarial assurance against the systems, tools, and authorization paths surrounding those agents.
Company record
Core
Agent identity · policy · MCP enforcement
Product
Averon IntelliScan
Service
Adversarial assessment · continuous assurance
Coverage
International
Research
Coordinated public disclosure
01 / Why Averon exists
Agents gained capability before they gained an authorization model.
AI systems now retrieve internal data, invoke tools, and initiate state-changing actions. Yet many still execute through one application identity and rely on natural-language instructions to limit behaviour.
Averon exists to move authority out of model context and into a control path: one identity per agent, explicit policy over tools, data, and actions, and enforcement before an MCP request reaches its target.
Operating thesis
The model should request authority, never define it.
- Identity must distinguish one agent from another.
- Permissions must be explicit and externally managed.
- Policy must live outside the prompt.
- Enforcement must sit in the request path.
02 / One security system
Prevention, detection, and evidence.
The product and assurance practices address different layers of the same problem. Their roles are stated clearly rather than collapsed into one claim.
02Detect
Averon IntelliScan
Prompt-injection and excessive-reasoning screening flags suspicious instructions and supports deeper escalation.
IntelliScan →03Test
Adversarial assurance
Assessment and continuous testing establish whether the deployed system and its surrounding controls hold under attack.
Assurance →03 / Assurance integrity
Product involvement is disclosed, not disguised.
Averon is a product company with an assurance capability. That relationship is material whenever an assessment includes Averon controls.
- 01Scope is explicitThe engagement states whether Averon products are inside the assessed boundary, outside it, or not present.
- 02Evidence leadsFindings require a reproduced path to data exposure, unauthorized action, or control failure—not a sales recommendation.
- 03Product adoption is not a findingA remediation describes the required security property. It does not require the client to purchase an Averon product.
- 04Conflicts are declaredAny product, vendor, or commercial relationship relevant to the assessment is stated before work begins.
- 05Severity follows methodologyFactual corrections are welcomed. Finding severity is not softened to protect a product or commercial relationship.
04 / Public research record
Authorization failure is not an abstract concern.
Averon research into cross-application authorization failures entered coordinated public disclosure through CERT/CC.
VU#849433 documented platform-level issues that enabled authenticated users to retrieve records belonging to other applications. The note was published on 08 July 2026 and credits Saud Darwish as reporter.
Research access
Anthropic Cyber Verification Program
Averon is a vetted CVP organization for legitimate defensive security research and penetration testing using select Claude models with reduced cyber safeguards.
Work with Averon
Start with the authority your agents hold today.
We can discuss the control platform, IntelliScan, an assurance engagement, or the relationship between them.