AI agent defense and server security guides
Learn what 1-SEC observes, which events it needs, how its checks are tested, and where another component must enforce the decision. Each guide links to the implementation and explains its coverage limits.
An alert is an observation. Prevention requires a working request gate or configured response with the right permissions and target context. Read our editorial and coverage policy.
AI agent containment
- AI Agent Goal Hijacking: Signals and Read-Only Goal Checks
- AI Agent Tool Policy: Blocked Tools and Sensitive Targets
- AI Agent Memory Poisoning and Cross-Session Signals
- Excessive Agency Monitoring: Scope, Approvals, Tools, and Plans
- AI Agent Network Egress: Web-Fetch Signals and Real Network Controls
- Agent Rapid Actions and Rogue Loops: Alert Thresholds, Not Hard Budgets
- AI Agent Spawn Depth: Reported Parent Chains and Monitoring
- Cascading Agent Failures and Retry Storm Monitoring
- AI Agent Approval Bypass Events and Trust Boundaries
- Shadow AI Monitoring: Known Endpoints and Reported Authorization
- AI Agent Scope Drift: Sensitive-Target History and Its Limits
Prompt injection and LLM firewall
AI agent identity and delegation
Self-hosted agent operations
Server and VPS defense
RAG and model security
Security operations and management access
MCP and tool security
- MCP Tool Description Poisoning: Local Pattern Inspection
- MCP Rug Pulls: Description Hash Changes and Reapproval
- Agent Tool Payloads: Shell, URI, and Destructive SQL Indicators
- MCP Cross-Server Tool Shadowing: Name Collision Signals
- MCP Remote Route Monitoring: Reported URLs and Hostname Checks
- Hidden Unicode Instructions: Instrumenting the LLM Input Scanner