AI asset inventory
Maintain an AI-BOM across models, agents, tools, data sources, deployments, guardrails and dependencies.
← Exposure ManagementPreviewAI-SPM · Built on Skyreach Orion
See what AI systems, agents and MCP tools can access—and what misuse could reach.
AI-SPM inventories AI applications, models, endpoints, agents, tools, MCP servers and knowledge sources. Orion connects them to existing identity, network, application and data relationships, giving teams visibility into effective agent permissions and potential blast radius without creating a disconnected AI inventory.
Evidence in
What Orion connects
The exact coverage depends on the integrations and authorised evidence available in your environment. Every record retains where it came from.
Example sources
- Azure AI and Microsoft Foundry services
- Amazon Bedrock and supported agent services
- Google Vertex AI and AI infrastructure
- Self-hosted model and application metadata
- Agent frameworks, traces and evaluation platforms
- MCP servers, tool declarations and identities
What the ontology models
- AI applications, models and endpoints
- Agents, tools and MCP servers
- Knowledge bases, vector stores and datasets
- Execution identities and effective permissions
- Guardrails, evaluations and approval controls
- Runtime workloads and software dependencies
Intelligence out
What visibility and insight you gain
The module uses the shared Orion model, so its findings can be understood through evidence collected by other security disciplines.
Agent permissions
Separate declared tool access from the effective permissions of the identity behind each tool.
MCP visibility
Surface MCP servers, exposed tools, authentication evidence and high-impact administrative actions.
AI blast radius
Show what infrastructure or sensitive data could be reached if an agent or integration were manipulated.
Connected analysis
An example of the exposure Orion can reveal
This is an illustrative relationship chain, not a claim that every path represents a confirmed compromise. Orion preserves evidence and confidence so teams can validate what is effective.
Untrusted prompt or content→AI agent→MCP tool identity→Sensitive system
Risks made visible
What the module is looking for
- Agents with excessive autonomy or privileged tools
- Unauthenticated or unapproved MCP servers
- Tool identities able to modify production infrastructure
- Knowledge bases containing sensitive information
- Missing human approval for destructive actions
- AI-originated paths to data exfiltration or privilege abuse
Questions it helps answer
Move from findings to decisions
- Which AI agents can change production?
- What identity does each MCP tool execute as?
- Which knowledge bases contain sensitive data?
- If this agent is manipulated, what could it reach or disclose?
Why Orion matters
This discipline does not operate in isolation
AI Security Posture Management contributes its evidence to the same ontology used by the rest of Exposure Management. That allows a finding to inherit network reachability, identity privilege, data sensitivity, software, application and business context where those facts exist.
See the complete Exposure Management productAI-SPM evidence+Shared ontology+Cross-domain context=Prioritised exposure
See what this means in your environment
Tell us which sources you use and the questions your teams struggle to answer today.