Inner Warden

Provides runtime security for AI agents that execute commands and access infrastructure, blocking prompt injection, tool poisoning, and malicious execution chains before they run. Sells to developers, IT and operations teams, and security teams in businesses that run autonomous coding and MCP-based agents on Linux hosts. Delivers as local-first software with kernel-enforced controls via eBPF/LSM, command and MCP screening, and tamper-evident audit trails without requiring a vendor cloud control plane.

Key features

• Command and MCP call screening
• Tool-poisoning and injection signals
• eBPF LSM kernel enforcement
• Agent-scoped execution gates
• Secret read gates
• DNS pre-authorization
• Sensitive-path read protection
• C2 and exfiltration signals
• Lateral-movement detection
• eBPF host telemetry and correlation
• Tamper-evident local decision history
• Signed off-host audit anchors
• Host agent tenant attribution
• 71 agent-threat rules
• 82 host detectors
• 47 eBPF programs for visibility