AI Agent Permission Risk Report
Cross-cloud and SaaS summary for a CISO, CTO, auditor, or board audience. Shows business impact, top paths, and first actions.
- Score
- 82 / 100 High
- Assessment ID
- KAV-ASM-2026-05-NSHC-001
- Best for
- Executive review
Synthetic product output
See what a Kavryl assessment delivers.
These sample reports use synthetic company, subscription, account, tenant, and resource IDs, but the format is modeled on Kavryl scanner output across Azure, AWS, GitHub, Microsoft 365 / Entra, Google Workspace, and MCP agent configs.
Downloadable PDFs
Board-ready reports, not raw scanner noise.
A Kavryl assessment turns read-only evidence into executive summary, technical findings, OWASP API Top 10 mapping, MITRE ATLAS-style AI risk paths, and owner-ready remediation steps.
Azure and AWS AI Agent Risk Report
Uses synthetic Azure subscription IDs, AWS account IDs, resource ARNs, and cloud findings based on our Azure and AWS scanners.
- Score
- 88 / 100 Critical path
- Systems
- Azure AI, Key Vault, Storage, AWS IAM, API Gateway
- Best for
- Cloud security
SaaS, OAuth, GitHub, and Agent Risk Report
Shows what Kavryl output looks like for Microsoft 365 / Entra, Google Workspace, GitHub, OAuth grants, and MCP tools.
- Score
- 76 / 100 High
- Systems
- M365, Entra, Google Workspace, GitHub, MCP
- Best for
- AppSec and IAM
Executive Summary
The report explains blast radius, not just misconfigurations.
Kavryl combines scanner evidence from cloud, SaaS, identity, repository, and agent configuration surfaces. The output tells security leaders which agent paths are dangerous, why they matter, and what owners should fix first.
Sample overall exposure score
82 / 100
High
Example Risk Path
A single finding may look manageable. The real risk appears when broad OAuth scopes, static secrets, public APIs, and privileged cloud roles combine into an agent action path.
Synthetic Evidence Examples
ID
System
Evidence
Interpretation
CLD-001
Azure
Service principal has Owner at subscription 4b812c7f-301d-45fb-8a4d-8a7750c4c19f.
Compromised agent identity could mutate cloud resources or grant access.
SAA-001
M365
Support Agent OAuth App has Mail.Read, Files.Read.All, and offline_access.
Delegated access can expose mail and files beyond the current workflow.
AWS-001
AWS
arn:aws:iam::391742608155:role/agent-bedrock-admin-role includes wildcard signal.
Agent or automation compromise could create broad cloud blast radius.
MCP-001
MCP
Filesystem and shell tools are available in the same support-agent profile.
Tool combination increases unsafe action and data movement risk.
Action Plan
First 7 days
Disable shell-capable MCP tools for support agents, require approval before outbound Slack/email actions, and pause new agent grants.
First 30 days
Replace broad Azure/AWS roles, remove unused OAuth scopes, split agent service identities, and restrict RAG data sources.
First 60 days
Run recurring scans, assign owners to every agent/tool, and create an approval workflow for new SaaS and cloud permissions.
Framework Mapping
Kavryl maps evidence to OWASP API Security Top 10 categories such as Broken Authentication, Broken Function Level Authorization, Security Misconfiguration, and Improper Inventory Management. It also maps AI-specific interpretation to MITRE ATLAS-style paths such as AI-enabled system discovery, credential misuse, excessive agency, and public agent/API exposure.
Kavryl Security