Security Overview

Posture across all connected agents and runtime activity.

Total AI Agents

142

Critical Findings

Policy Coverage

86%

Runtime Calls

12,480

Risk distribution

critical (7)high (14)medium (28)low (41)

Runtime activity (7 days)

Findings by category

Lethal Trifecta exposure

2 agents ungated

Agents that can ingest untrusted input, reach external systems, and take irreversible actions. All three legs with no gate = lethal.

deploy-botLethal Trifecta

support-botTrifecta · gated

invoice-agentLethal Trifecta

data-indexerPartial

pr-reviewerPartial

Top risks

Agent can attach IAM policies in production

support-bot · PE3

critical94

Deploy agent executes unpinned remote scripts

deploy-bot · SC2

critical90

Environment variable harvesting in MCP tool

support-bot · E2

high81

Known vulnerable dependency (CVE via OSV.dev)

data-indexer · SC4

high76

MCP tool requests excessive privileges

invoice-agent · MCP1

high73

Connector health

GitHubhealthy

AWShealthy

Slackhealthy

MCPdegraded