Aeon AI Risk Management

Penetration testing for systems that can access, execute, or leak.

Aeon CyberGuard is an AI-native pentest lab. It starts with a free public-exposure snapshot and free threat-model call, then moves to a paid diagnostic or signed 10-business-day AI-native pentest review, and finally to an ongoing Release Security Gate retainer.

Questions this page answers

What does the CyberGuard AI-native pentest lab cover?
CyberGuard reviews what websites, applications, APIs, AI agents, MCP servers, RAG systems, tool calls, and AI/ML platforms can access, execute, or leak across authorization, command boundaries, and data paths.
What is the first tier?
The free public-exposure snapshot uses public sources only to map externally visible AI attack-surface signals. It does not test, probe, or access prospect systems.
Is CyberGuard authorized-only?
Yes. The free snapshot and threat-model call are passive and public-only. Any diagnostic, Review, or Release Security Gate testing starts only after signed Rules of Engagement, scope, named authorization owner, test window, data handling, and stop conditions.
What are the CyberGuard steps from first look to ongoing coverage?
Start with a free public-exposure snapshot built from public data, review it in a free 60-minute threat-model call, then choose a paid diagnostic or full 10-business-day AI-native pentest review under signed ROE. Teams shipping AI agents continuously can move into the ongoing Release Security Gate retainer.
Is CyberGuard backed by active security research?
Yes. Aeon's owned-lab, AI-driven, human-reviewed pipeline found 101 authorization vulnerabilities across 60 open-source AI and infrastructure projects, including 4 Critical and 32 High findings. Aeon's verified public track record includes seven patched security advisories and four assigned CVEs.
Does Aeon provide SOC 2 certification?
No. Aeon supports SOC 2 readiness with audit-ready evidence your assessor can map to your controls. Your auditor still issues the opinion. Aeon does not issue SOC 2 reports, ISO certifications, or assurance opinions. Independent CVE Numbering Authorities assign CVE identifiers.

Cybersecurity operations powered by Claude Mythos 5.1.

Written scope comes first. AI-assisted investigation is followed by specialist validation, remediation priorities and retest evidence. Model access is an enabling capability, not a guarantee that every vulnerability will be found.

Access

Test whether one user, tenant, or role can reach another's data through agents, APIs, RAG collections, model-platform objects, or support tools.

Execute

Review whether an agent, MCP server, or tool call can be driven to perform unauthorized actions, including indirect prompt injection, command-boundary issues, and unsafe internal API reach.

Leak

Follow data paths across AI integrations, SSRF exposure, secret handling, unsafe retrieval flows, logs, and customer-security evidence.

Free public-exposure snapshot

The first tier maps externally visible AI attack-surface signals from public sources only, with no access, probing, legal paperwork, or vulnerability claim.

Free threat-model call

A 60-minute call explains what the snapshot means, the single top fix, and what would require signed authorization.

Release Security Gate retainer

Ongoing coverage for major agent, model, MCP, AI platform, or configuration releases, with verified remediation and retest evidence.

SOC 2 readiness evidence

Findings are mapped to OWASP LLM Top 10, MITRE ATLAS, and your SOC 2 and NIST AI RMF controls so your assessor can use the evidence while your auditor still issues the opinion.

Research-backed proof

Aeon's owned-lab, AI-driven, human-reviewed pipeline found 101 authorization vulnerabilities across 60 open-source AI and infrastructure projects. The separate verified public record contains seven patched advisories and four assigned CVEs.

Public proof assets

The frozen research evidence snapshot, AI Exposure Assessment sample, CyberGuard one-pager, and executive briefing deck are public. The evidence-package sample and ROE template show how the AI-native pentest lab scopes, validates, fixes, and retests findings.

Proof boundary

Aeon is firm-attributed, authorized-only, human-verified, and AI-assisted. Seven coordinated disclosure records are public and patched, including four assigned CVEs. Other private records remain aggregate-only.

What we will clarify on the call

Bring one application, API or AI-agent workflow to a 30-minute scoping call. We will clarify the authorization boundary, evidence needed and proposed assessment deliverables, with expert validation at the core. The system, business risk and authorization boundaries to review. The evidence and access needed for a proposed assessment, without sharing secrets on the call. A proposed scope, deliverables and next commercial step. The call is not a free vulnerability assessment.

What a validated finding should contain

Synthetic example only, not a client finding or a Mythos benchmark. Illustrative scenario: cross-tenant document access in a fictional test application. No live customer system was tested for this example. Boundary and impact: An account in Tenant A can retrieve a synthetic document assigned to Tenant B. The proposed impact is cross-tenant data exposure, subject to validation in the agreed environment. Validation evidence: A redacted request-and-response comparison using authorized test accounts and synthetic data, plus affected components and reproduction conditions. Remediation: Enforce server-side object and tenant authorization for each document request. Do not rely on hidden interface controls. Retest and limitations: Repeat the cross-tenant negative test and legitimate-access regression checks. Record the result, tested scope and remaining uncertainty, not a guarantee that all vulnerabilities have been found.