# CyberFortify Guardian Agent Security

> Autonomous cyber defense for the agent-operated Internet. Guardian detects, decides, contains, recovers, and proves at machine speed inside signed policy.

## Authoritative pages
- Agent Security: https://www.cyberfortifysolutions.com/gaas#agent-security
- Agent Trust Center: https://agents.cyberfortifysolutions.com
- Machine-readable trust manifest: https://agents.cyberfortifysolutions.com/api/agent-trust
- Public read-only assessment contract: https://agents.cyberfortifysolutions.com/api/agent-assessment
- A2A Agent Card: https://agents.cyberfortifysolutions.com/.well-known/agent-card.json
- Read-only MCP endpoint: https://agents.cyberfortifysolutions.com/mcp
- MCP Server Card: https://agents.cyberfortifysolutions.com/.well-known/mcp/server-card.json
- Agent Skills discovery: https://agents.cyberfortifysolutions.com/.well-known/agent-skills/index.json
- CyberFortify Labs: https://www.cyberfortifysolutions.com/labs

## Capabilities
- Discover: Inventory sanctioned and shadow agents, MCP services, browser automation, and delegated identities.
- Decide: Continuously evaluate identity, tenant, tool, target, risk, and signed mission policy at machine speed.
- Defend: Detect and disrupt malicious activity through a continuous, policy-bound defensive workflow.
- Contain: Revoke access, terminate sessions, disable tools, or isolate the affected endpoint when needed.
- Prove: Preserve a readable evidence chain showing who authorized what, what ran, and what changed.
- Recover: Validate containment, restore the safe operating state, and keep watching for recurrence automatically.

## Security principles
- Sponsor before access: A sponsor-registered workload credential must be attributable to an authorized sponsor and authorization context before access.
- Least authority: Every tool and resource is denied by default, narrowly scoped, and available only for the current task.
- Policy, not permission queues: Humans establish signed mission policy and break-glass authority. Guardian acts autonomously inside it and fails closed outside it.
- Tenant isolation: The sponsor-registered workload credential, data access, approval, and action scope remain bound to the same authorized organization.
- Verifiable actions: Security decisions and resulting actions produce evidence that operators can inspect and audit.
- Rapid revocation: Agent sessions, credentials, and tools must be independently revocable without disabling the human owner.

## Public-surface boundary
The public agent endpoint is informational and read-only. It can perform bounded static checks of submitted Agent Cards, MCP server cards, tool schemas, and configurations. It does not contact external targets, expose customer data, use credentials, install software, execute Guardian operational tools, or provide write access. Payment never grants authority.

## Operating model
Humans define signed mission policy, scope, risk tolerance, and break-glass authority. Guardian performs routine defensive operations autonomously. Work outside delegated policy fails closed.

## Contact
Agent Security Readiness Assessment: info@cyberfortifysolutions.com