Skip to content

SCAIROS AgentScope

See what your AI agent can access, expose and execute

Submit a sanitized architecture description. AgentScope maps identities, tools, data paths and permissions, then calculates the operational blast radius using deterministic security scoring.

✓Architecture topology
✓Attack-path analysis
✓Permission heatmap
✓Deterministic risk scoring
✓OWASP and NIST mappings
✓30/60/90-day remediation

Architecture input

Analyze your AI agent

Include the agent’s users, model, RAG system, databases, credentials, tools, actions, approvals, tenant boundaries, limits, logs and shutdown controls.

Minimum 100 characters0 / 19,999

What AgentScope analyzes

Identity and credentials

Shared accounts, broad roles, administrative authority and missing workload identities.

Tool authorization

Actions, resources, transaction limits, destinations, approval gates and reversibility.

Data and RAG isolation

Sensitive information, retrieval filters, authenticated tenant boundaries and export paths.

Prompt and input security

Direct and indirect prompt injection through users, documents, email, websites and retrieval.

Evidence and monitoring

Identity, tenant, retrieval, model, policy, tool, approval and outcome audit records.

Containment

Kill switches, credential revocation, rate limits, spending limits and incident recovery.

AgentScope is a defensive architecture-analysis tool. Results are based on the submitted description and are not a penetration test, certification or guarantee of security.