The ten risks specific to AI agents, from goal hijack to rogue agents, and the live control that meets each one.
Last verified: 30 October 2026
Every row is a control that is live in AxiSentinel or AxiReady today, with the phase it belongs to.
| Requirement | What AxiSentinel and AxiReady do | Phases |
|---|---|---|
| ASI01 Agent Goal Hijack | Injection cases in the agentic test suite; outbound and high-impact tools gated; behavior guard flags a change of pattern. | 03, 05, 07, 10 |
| ASI02 Tool Misuse and Exploitation | MCP Gateway discovers and gates every tool server; unsafe call sequences are held at the authority gate; unapproved tools blocked. | 03, 05, 06 |
| ASI03 Identity and Privilege Abuse | Named identity, role scope and credential reference on every action; no shared service accounts; quarantine at the gateway. | 04, 08 |
| ASI04 Agentic Supply Chain | Discovery of frameworks, connectors, MCP servers and model files; AI-BOM with CycloneDX, SPDX, Trivy, Grype and OSV.dev; approval before first use. | 01, 03 |
| ASI05 Unexpected Code Execution | Code-execution tools gated and approved per agent; authorized adversarial tests after every change. | 03, 10 |
| ASI06 Memory and Context Poisoning | Behavior guard learns each agent's normal and flags drift; re-tests after every model, data or vendor change; retrieval sources in the inventory. | 07, 10 |
| ASI07 Insecure Inter-Agent Communication | Agent-to-agent calls pass the gateway under each agent's own identity; anomalies quarantined. | 04, 05, 08 |
| ASI08 Cascading Failures | Quarantine isolates one agent without stopping the fleet; the kill switch stops it on one confirmed decision; incidents carry their clock. | 08, 09 |
| ASI09 Human-Agent Trust Exploitation | High-impact actions wait for a named person in the approvals inbox, with the reason recorded; summaries tested in the suite. | 06, 10 |
| ASI10 Rogue Agents | Drift, spikes and signs of hijack flagged on the AI radar; Stop recommended at 10x normal; stop and revoke built in. | 07, 08 |
Sources: OWASP GenAI Security Project, Top 10 for Agentic Applications 2026; Cloud Security Alliance and Token Security survey, April 2026; frontier-lab incident disclosures, July to October 2026.
AxiReady finds. AxiSentinel gates, holds, quarantines and stops, Always On or On Demand. A qualified AxiLayer AI auditor signs against OWASP Agentic Top 10.
Twelve questions, three minutes, a band and three first steps. Then the 48-Hour Readiness Report for your estate.
